Fishing boat activity recognition and prediction method based on multi-source data fusion and physical constraints

Abstract

This application provides a method for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints, belonging to the field of data processing. It acquires AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data in the target sea area; extracts ground speed and heading change characteristics from the AIS trajectory data, and constructs a temporal state vector through temporal delay embedding to achieve identification of fishing and navigation states; identifies covert nighttime operation targets under AIS silent conditions by combining nighttime light remote sensing data; constructs a fishing vessel activity intensity field by fusing fishing activities and covert operation activities, and forms a continuous spatiotemporal sequence sample by combining marine environmental data; and uses a physically constrained ConvLSTM-PINN prediction model to achieve spatiotemporal prediction of fishing vessel activity intensity. Implementing this technical solution enables accurate identification of the actual operational behavior of fishing vessels and spatiotemporal prediction of activities conforming to marine physical laws.

Description

Technical Field

[0001] This application relates to the technical field of data processing, specifically to a method for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints. Background Technology

[0002] With the increasing intensity of global marine resource development and utilization, fishing activities play a vital role in ensuring food security and promoting economic development in coastal areas. At the same time, the spatial and temporal distribution of fishing activities is influenced by multiple factors, including the natural environment, resource conditions, and management systems, resulting in significant fluctuations and randomness in the activity process, posing challenges to relevant departments in supervising fishing activities.

[0003] Currently, relevant technologies mainly rely on AIS trajectory data to identify the activity status of fishing vessels. However, in actual marine surveillance scenarios, some fishing vessels may experience trajectory silence due to equipment failure, limited communication conditions, or deliberate shutdown of AIS equipment, resulting in long-term missing AIS trajectory data. This leads to discrepancies between the monitoring results of fishing vessel activities and the actual operational situation, affecting the accuracy of subsequent regulatory decisions. At the same time, most existing fishing vessel activity prediction methods rely on learning statistical patterns from historical data, lacking constraints from physical laws such as ocean currents, wind fields, and marine ecology. When the marine environment undergoes complex changes or historical samples are missing, noisy, or have distribution shifts, the prediction model is prone to producing prediction results that do not match the actual marine environmental conditions.

[0004] Therefore, how to accurately identify the actual operational behavior of fishing vessels and predict their activities in accordance with the laws of ocean physics under conditions of silent AIS tracks and complex changes in the marine environment has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a method for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints. This method facilitates the accurate identification of the actual operational behavior of fishing vessels and the spatiotemporal prediction of activities that conform to the laws of marine physics, even under conditions where there are silent gaps in AIS trajectories and complex changes in the marine environment.

[0006] The first aspect of this application provides a method for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints. The method includes: acquiring AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data in a target sea area; extracting ground speed features and heading change features from the AIS trajectory data, and performing time-series delay embedding processing based on the ground speed features and heading change features to construct a time-series state vector; performing cluster analysis on the time-series state vector to form fishing state categories and navigation state categories, and performing state labeling on the AIS trajectory data based on the fishing state categories and navigation state categories to form fishing vessel activity trajectory data; and performing state labeling on the VIIRS / DNB nighttime light remote sensing data... DNB nighttime light remote sensing data and the fishing vessel activity trajectory data are subjected to spatiotemporal cross-matching processing to identify unmatched nighttime light targets. Based on the radiance distribution of the unmatched nighttime light targets, concealed nighttime operation targets are identified to form abnormal fishing vessel activity identification results. Based on the abnormal fishing vessel activity identification results and the fishing vessel activity trajectory data corresponding to the fishing status category, a fishing vessel activity intensity field is constructed, and combined with the marine environmental data to form a continuous spatiotemporal sequence sample. The continuous spatiotemporal sequence sample is input into the ConvLSTM-PINN prediction model to generate a fishing vessel activity intensity prediction result based on a physical constraint mechanism constructed from ocean current constraint information, spatial continuity constraint information, wind field safety constraint information, and biological coupling constraint information.

[0007] A second aspect of this application provides a device for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints. The device includes an acquisition module and a processing module. The acquisition module is used to acquire AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data in a target sea area. The processing module is used to extract ground speed features and heading change features from the AIS trajectory data, and perform time-delay embedding processing based on the ground speed features and heading change features to construct a time-series state vector. The processing module is further used to perform cluster analysis on the time-series state vector to form fishing state categories and navigation state categories, and perform state labeling on the AIS trajectory data based on the fishing state categories and navigation state categories to form fishing vessel activity trajectory data. The module is further configured to perform spatiotemporal cross-matching processing on the VIIRS / DNB nighttime light remote sensing data and the fishing vessel activity trajectory data to identify unmatched nighttime light targets, and to identify concealed nighttime operation targets based on the radiance distribution of the unmatched nighttime light targets, thereby forming an abnormal fishing vessel activity identification result; the processing module is further configured to construct a fishing vessel activity intensity field based on the abnormal fishing vessel activity identification result and the fishing vessel activity trajectory data corresponding to the fishing status category, and to form a continuous spatiotemporal sequence sample by combining the marine environmental data; the processing module is further configured to input the continuous spatiotemporal sequence sample into the ConvLSTM-PINN prediction model to generate a fishing vessel activity intensity prediction result based on a physical constraint mechanism constructed from ocean current constraint information, spatial continuity constraint information, wind field safety constraint information, and biological coupling constraint information.

[0008] A third aspect of this application provides an electronic device comprising a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the fishing vessel activity identification and prediction method based on multi-source data fusion and physical constraints as described above.

[0009] In a fourth aspect, this application provides a non-transitory computer-readable storage medium storing instructions that, when executed, perform the fishing vessel activity identification and prediction method based on multi-source data fusion and physical constraints as described above.

[0010] In summary, one or more technical solutions provided in this application have at least the following technical effects or advantages: By integrating AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data, this system achieves integrated processing for fishing vessel activity identification and spatiotemporal prediction. First, a temporal state vector is constructed using the ground speed and heading change characteristics from the AIS trajectory. Cluster analysis is then used to automatically identify fishing and navigation states, improving the accuracy of fishing vessel behavior assessment. Subsequently, VIIRS / DNB nighttime light remote sensing data is introduced and spatiotemporally cross-matched with the fishing vessel activity trajectory data. By identifying high-brightness nighttime operating targets among the unmatched nighttime light targets, effective detection of covert fishing activities under silent AIS trajectory conditions is achieved, compensating for monitoring blind spots inherent in single AIS data sources. Furthermore, the results of abnormal fishing vessel activity identification are combined with the fishing vessel activity trajectory data corresponding to the fishing state category to construct a fishing vessel activity intensity field. This field is then combined with marine environmental data such as wind field, ocean current, and chlorophyll concentration to form a continuous spatiotemporal sequence sample. Finally, the ConvLSTM-PINN prediction model was used to predict the intensity of the activity. A physical constraint mechanism was constructed through ocean current constraints, spatial continuity constraints, wind field safety constraints, and biological coupling constraints. This ensures that the prediction results not only reflect historical activity patterns but also conform to marine physical and ecological laws, thereby improving the completeness of fishing vessel activity identification and the accuracy and reliability of spatiotemporal prediction of activities. Attached Figure Description

[0011] Figure 1 A flowchart illustrating the fishing vessel activity identification and prediction method based on multi-source data fusion and physical constraints provided in this application embodiment; Figure 2 A schematic diagram of a module for a fishing vessel activity identification and prediction device based on multi-source data fusion and physical constraints provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0012] Explanation of reference numerals in the attached figures: 21. Acquisition module; 22. Processing module; 31. Processor; 32. Communication bus; 33. User interface; 34. Network interface; 35. Memory. Detailed Implementation

[0013] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0014] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0015] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. In addition, the terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0016] To address the aforementioned technical problems, this application provides a method for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the fishing vessel activity identification and prediction method based on multi-source data fusion and physical constraints provided in this application embodiment. The method is applied to a server and includes steps S110 to S160, the specific steps of which are as follows:

[0017] S110: Acquire AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data in the target sea area.

[0018] Specifically, the server first receives the spatial range and target time period corresponding to the target sea area. The spatial range can be determined by the minimum and maximum longitude, minimum and maximum latitude, and the target time period can be determined by the start and end times. The target sea area is the common geographic object for all subsequent data filtering and alignment, and the target time period is the common temporal object for all subsequent data filtering and alignment. When the server subsequently acquires raw AIS trajectory data, raw VIIRS / DNB nighttime light remote sensing data, and raw marine environmental data, it uses the spatial range and target time period as unified constraints to ensure that the three types of data describe the same sea area and the same time process, avoiding spatial or temporal mismatches in subsequent state identification, nighttime light matching, and spatiotemporal prediction due to inconsistent data coverage.

[0019] The server acquires raw AIS trajectory data from the target sea area. This raw AIS trajectory data is a dynamic record of vessels broadcast by the Automatic Identification System (AIS), and includes at least the vessel identifier, vessel type information, trajectory time, latitude and longitude, speed relative to the ground, and heading angle. The vessel identifier distinguishes different vessels, the vessel type information determines whether the corresponding vessel is a fishing vessel, the trajectory time determines the order of trajectory points, the latitude and longitude determines the vessel's spatial position in the target sea area, the speed relative to the ground represents the vessel's speed, and the heading angle represents the vessel's direction of motion. The server filters fishing vessel trajectory points based on the vessel type information and eliminates non-fishing vessel trajectory points, ensuring that subsequent processing focuses solely on fishing vessel activities. Then, it assigns fishing vessel trajectory points to the same vessel identifier, grouping them into the same trajectory set. Finally, it sorts the fishing vessel trajectory points within the same trajectory set in ascending order based on the trajectory time, thus forming a continuous AIS trajectory sequence.

[0020] The server performs validity checks on continuous AIS trajectory sequences. Validity checks determine whether each trajectory point meets the requirements for spatial rationality, temporal rationality, speed rationality, and field completeness. Spatial rationality means the latitude and longitude of the trajectory point should fall within a specified spatial range; temporal rationality means the trajectory time should fall within the target time period; speed rationality means the trajectory speed above ground should not exceed the reasonable speed range achievable by the fishing vessel; and field completeness means that key fields such as vessel identification, trajectory time, latitude and longitude, speed above ground, and heading angle should not be empty. The server identifies trajectory points that do not meet the above conditions as abnormal trajectory points and removes them from the continuous AIS trajectory sequence to reduce interference from equipment false alarms, positioning drift, and missing data on subsequent fishing vessel status identification.

[0021] After removing abnormal trajectory points, the server further determines whether there are any missing trajectory points between adjacent trajectory points. Missing trajectory points refer to intermediate trajectory points in a continuous AIS trajectory sequence corresponding to the same ship identifier that were not recorded due to short-term AIS signal loss, communication obstruction, or reception delay. The server judges the degree of missing data based on the time interval between adjacent trajectory points. When the time interval is less than or equal to a preset completion time threshold, the corresponding trajectory segment is determined to be a short-term missing trajectory segment, and the missing trajectory points are completed based on the trajectory time, latitude and longitude, ground speed, and heading angle of adjacent trajectory points. When the time interval is greater than the preset completion time threshold, the corresponding trajectory segment is determined to be a trajectory interruption segment, and the trajectory interruption mark is retained without forced completion to avoid mistakenly correcting potential AIS silent behavior as a normal trajectory. The expression for calculating the time interval is:

[0022]

[0023] in, Indicates the first number corresponding to the same ship identification The trajectory point and the first The time interval between trajectory points Indicates the first The trajectory time of each trajectory point Indicates the first The trajectory time of each trajectory point This represents the sequential number of trajectory points within the same continuous AIS trajectory sequence, ranging from the first trajectory point to the second-to-last trajectory point. This expression reflects the continuity of the AIS trajectory through the trajectory time difference between adjacent trajectory points. The server compares these values. It distinguishes between short-term missing trajectory segments and trajectory interruption segments by comparing them with the preset completion time threshold.

[0024] When the server completes short-term missing trajectory segments, it can interpolate the latitude and longitude positions of the missing trajectory points based on the time ratio between adjacent trajectory points. The expression for the interpolated position is:

[0025] in, Indicates the missing trajectory points in the trajectory time. Interpolation position at that point Indicates the first The latitude and longitude positions of the valid trajectory points Indicates the first The latitude and longitude positions of the valid trajectory points This indicates the time required to complete the missing trajectory points. Indicates the first The trajectory time of each valid trajectory point Indicates the first The expression determines the transition position of a missing trajectory point in space based on the time ratio between two valid trajectory points, ensuring that the completed trajectory point maintains temporal and spatial continuity with the preceding and following valid trajectory points. After the server completes anomaly removal, short-term completion, and interruption marking, it generates target AIS trajectory data. This target AIS trajectory data is used for subsequent extraction of ground speed characteristics, heading change characteristics, and trajectory interruption status.

[0026] The server acquires raw VIIRS / DNB nighttime light remote sensing data from the target sea area. This raw VIIRS / DNB nighttime light remote sensing data is nighttime sea surface light detection data acquired by a visible infrared imaging radiometer in the day and night bands. It includes at least the nighttime light detection time, nighttime light detection location, radiance value, detection confidence level, and quality label. The nighttime light detection time determines the time of appearance of the nighttime light target; the nighttime light detection location determines the spatial location of the nighttime light target; the radiance value characterizes the light radiation intensity of the nighttime light target; the detection confidence level characterizes the degree to which the nighttime light target was reliably detected; and the quality label indicates whether the nighttime light detection point is affected by clouds, moonlight, shoreline lights, sensor malfunctions, or other background noise. The server trims the raw VIIRS / DNB nighttime light remote sensing data according to the spatial range and target time period, retaining nighttime light detection points located within the target sea area and target time period. Then, invalid nighttime light detection points are removed based on the quality label, and those meeting the detection confidence level requirements are retained to form the target nighttime light detection point set. The target nighttime light detection point set is used for subsequent spatiotemporal cross-matching with fishing vessel activity trajectory data to discover nighttime light targets lacking AIS trajectory support.

[0027] The server acquires raw marine environmental data from the target sea area, including wind field data, ocean current data, and chlorophyll concentration data. Wind field data characterizes the distribution of wind speed and direction in the target sea area, reflecting whether fishing vessels have safe operating conditions. Ocean current data characterizes the direction and intensity of seawater flow in the target sea area, reflecting the conditions for the convergence or diffusion of fish schools, plankton, and nutrients. Chlorophyll concentration data characterizes the distribution of primary productivity in the target sea area, reflecting the ecological basis that may attract fish schools. The server performs cropping processing on the wind field data, ocean current data, and chlorophyll concentration data based on spatial range and target time period, ensuring that the environmental data retains only the valid portion within the target sea area and target time period. Subsequently, missing data in the wind field data, ocean current data, and chlorophyll concentration data are filled in to form an environmental dataset. For data with localized spatial missing values, the server can use environmental values ​​from adjacent spatial locations to fill in the gaps; for data with short-term temporal missing values, the server can use environmental values ​​from adjacent time periods to fill in the gaps, so that the environmental data set has a continuously usable environmental description capability during subsequent spatial and temporal alignment processes.

[0028] The server performs spatiotemporal alignment processing on the target AIS trajectory data, the target nighttime light detection point set, and the environmental data set based on a unified spatiotemporal reference. The unified spatiotemporal reference includes a unified spatial grid and a unified time step. The unified spatial grid is used to unify data with different spatial representations into the same spatial unit, and the unified time step is used to unify data with different sampling frequencies into the same time unit. For the target AIS trajectory data, the server maps the trajectory points to corresponding spatial units based on their latitude and longitude, and to corresponding time units based on their trajectory time. For the target nighttime light detection point set, the server maps the nighttime light detection points to corresponding spatial units based on their nighttime light detection location, and to corresponding time units based on their nighttime light detection time. For the environmental data set, the server maps wind field data, ocean current data, and chlorophyll concentration data to corresponding spatial and time units based on the environmental raster position and environmental recording time. The formula for determining the spatial unit index is:

[0029]

[0030] in, Indicates latitude and longitude location Corresponding spatial unit index, Indicates the longitude of the data point to be mapped. Indicates the latitude of the data point to be mapped. Indicates the minimum longitude of the target sea area. Indicates the minimum latitude of the target sea area. This indicates the grid spacing of the unified spatial grid along the longitude direction. This indicates the grid spacing of the unified spatial grid in the latitudinal direction. This indicates rounding down. This expression determines the spatial cell into which the data point falls by dividing the spatial offset of the data point relative to the lower left boundary of the target sea area by the grid interval, so that AIS trajectory points, night light detection points, and marine environmental data can be placed in the same spatial grid system.

[0031] The formula for determining the time unit index is:

[0032] in, This indicates the time unit index corresponding to the trajectory time, night light detection time, or environmental recording time. Indicates the time of the data to be mapped. Indicates the start time of the target time period. Indicates a uniform time step. This indicates rounding down. The expression determines the time unit to which the data belongs by dividing the offset of the time of the data to be mapped relative to the start time of the target time period by a uniform time step, enabling data from different sampling frequencies to undergo subsequent fusion processing on the same time scale.

[0033] After completing the spatiotemporal alignment process, the server establishes a data association index. This index records the target AIS trajectory data, target nighttime light detection point set, and environmental data set within the same spatial and temporal unit. Specifically, when target AIS trajectory data exists within a spatial and temporal unit, the index records the corresponding ship identifier, trajectory time, latitude and longitude position, ground speed, heading angle, and trajectory interruption marker. When the target nighttime light detection point set exists within the same spatial and temporal unit, the index records the corresponding nighttime light detection time, nighttime light detection location, radiance value, detection confidence level, and quality indicator. When the environmental data set exists within the same spatial and temporal unit, the index records the corresponding wind field data, ocean current data, and chlorophyll concentration data. The resulting AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data are not isolated data sets, but rather a data foundation established through unified spatial units, unified temporal units, and data association indexes. This foundation supports subsequent construction of temporal state vectors, spatiotemporal cross-matching of nighttime light targets, construction of fishing vessel activity intensity fields, and generation of continuous spatiotemporal sequence samples.

[0034] S120. Extract ground speed features and heading change features from AIS trajectory data, and perform time-delay embedding processing based on ground speed features and heading change features to construct a time-series state vector.

[0035] Specifically, the server first reads the previously formed continuous AIS trajectory sequence. A continuous AIS trajectory sequence refers to a data sequence formed by arranging multiple trajectory points corresponding to the same vessel identifier in chronological order. Each trajectory point includes at least trajectory time, latitude and longitude position, ground speed, and heading angle. Ground speed is the speed of the fishing vessel relative to the ground recorded by the AIS device, reflecting the speed of the fishing vessel at the corresponding trajectory point. The server extracts the ground speed of each trajectory point sequentially according to the trajectory time and uses this ground speed as the initial ground speed feature. Simultaneously, it verifies the reasonableness of the initial ground speed feature by combining the changes in latitude and longitude position and trajectory time between adjacent trajectory points. When the difference between the initial ground speed feature and the reference ground speed calculated from adjacent trajectory points exceeds a preset speed difference threshold, or when the initial ground speed feature exceeds the reasonable speed range of the fishing vessel, the initial ground speed feature is identified as an abnormal ground speed. It is then corrected using the ground speeds corresponding to adjacent valid trajectory points, thus forming the target ground speed feature. The target's ground speed characteristic retains the actual speed record from the AIS trajectory data while reducing the impact of equipment drift, signal anomalies, or data mutations on subsequent status identification. The reference formula for calculating ground speed is:

[0036]

[0037] in, Indicates the first The trajectory point and the first The reference ground speed between the trajectory points is used to verify the first... Are the initial ground speed characteristics corresponding to each trajectory point reasonable? Indicates the first The latitude and longitude positions of the trajectory points are specifically determined by the first... The trajectory consists of the longitude and latitude of each trajectory point, which the server reads directly from a continuous AIS trajectory sequence; Indicates the first The server directly reads the latitude and longitude positions of each trajectory point from a continuous AIS trajectory sequence; Indicates the first The trajectory point and the first The spatial distance between trajectory points is calculated from the latitude and longitude positions of the two trajectory points; Indicates the first The trajectory time of each trajectory point is read by the server from a continuous AIS trajectory sequence; Indicates the first The trajectory time of each trajectory point is read by the server from a continuous AIS trajectory sequence. This expression calculates the average speed of the fishing boat within the adjacent trajectory segment by dividing the spatial distance between adjacent trajectory points by the time difference between adjacent trajectory points, and uses this as a reference value to verify whether the speed recorded by AIS is abnormal.

[0038] After identifying abnormal ground speeds, the server corrects them. This correction doesn't simply delete the abnormal trajectory points; instead, it restores reasonable speed values ​​based on the speed change trends of adjacent valid trajectory points within the same continuous AIS trajectory sequence. When both adjacent valid trajectory points exhibit target ground speed characteristics, the server interpolates and corrects based on the time ratio of the abnormal trajectory point between its preceding and following valid trajectory points. When only one valid trajectory point exists ahead or behind, the server uses the target ground speed characteristics corresponding to the nearest valid trajectory point for preservation correction. When multiple consecutive trajectory points exhibit abnormal ground speeds and cannot be corrected through short-time interpolation, the server retains the abnormal marker to avoid forcibly using unreliable speeds in subsequent time-series state vector construction. The interpolation correction expression is:

[0039]

[0040] in, Indicates the abnormal trajectory points in the trajectory time. The corrected target ground speed characteristics; This indicates the target's ground speed characteristics corresponding to the nearest valid trajectory point before the abnormal trajectory point, which the server obtains by forward searching through a continuous AIS trajectory sequence. This indicates the target's ground speed characteristic corresponding to the nearest valid trajectory point after the abnormal trajectory point, which the server obtains by searching through a continuous AIS trajectory sequence. The trajectory time of the abnormal trajectory point is read by the server from a continuous AIS trajectory sequence; This indicates the trajectory time of the nearest valid trajectory point before the abnormal trajectory point, which the server obtains by forward searching through a continuous AIS trajectory sequence. This represents the trajectory time of the nearest valid trajectory point after the anomalous trajectory point, obtained by the server through forward retrieval of the continuous AIS trajectory sequence. This expression reconstructs the target's ground speed characteristics consistent with the adjacent velocity change trends, based on the temporal position of the anomalous trajectory point between its preceding and following valid trajectory points, thus maintaining smooth and interpretable velocity changes in the continuous AIS trajectory sequence.

[0041] The server then extracts heading change features based on the heading angles in the continuous AIS trajectory sequence. The heading angle is the direction of the fishing vessel's movement recorded in the AIS trajectory data, typically taken as a reference north and clockwise, ranging from 0 to 360 degrees. The degree of heading change refers to the magnitude of the change in the fishing vessel's direction of movement between adjacent trajectory points, used to reflect whether the fishing vessel is frequently turning, circling, or maneuvering. Because the heading angle is periodic, when the heading angle changes from 359 degrees to 2 degrees, the actual degree of change should be 3 degrees, not 357 degrees. Therefore, the server needs to perform a periodic correction when calculating the degree of heading change. The expression for calculating the degree of heading change is:

[0042]

[0043] in, Indicates the first The trajectory point and the first The degree of change in heading between each trajectory point; Indicates the first The heading angle of each trajectory point is read by the server from a continuous AIS trajectory sequence; Indicates the first The heading angle of each trajectory point is read by the server from a continuous AIS trajectory sequence; This represents the original heading angle difference between adjacent trajectory points; This represents the compensated heading angle difference value when crossing the 0-degree and 360-degree boundaries. This expression obtains the minimum heading change degree that conforms to the actual turning situation by taking the smaller value between the original heading angle difference value and the compensated heading angle difference value, thus avoiding the turning amplitude being erroneously amplified due to the angular period boundary.

[0044] The server further combines the time intervals between corresponding trajectory points to form heading change features. These features represent the rate of heading change per unit time. Compared to the simple degree of heading change, these features eliminate the impact of inconsistent sampling intervals between different trajectory points. For example, a large directional change within a short period usually corresponds to frequent maneuvers, while a similar directional change over a longer period does not necessarily indicate high-frequency turning. The server determines the heading change features by the ratio of the degree of heading change between adjacent trajectory points to the corresponding time interval, specifically expressed as:

[0045]

[0046] in, Indicates the first The heading change characteristics corresponding to each trajectory point; Indicates the first The trajectory point and the first The degree of heading change between each trajectory point is obtained through the calculation expression of the degree of heading change; Indicates the first The trajectory time of each trajectory point is read by the server from a continuous AIS trajectory sequence; Indicates the first The trajectory time of each trajectory point is read by the server from a continuous AIS trajectory sequence. This expression obtains the turning intensity per unit time by distributing the degree of heading change to the time interval between adjacent trajectory points, enabling stable differentiation between low-speed, high-frequency turning characteristics in fishing conditions and high-speed, low-frequency turning characteristics in navigation conditions.

[0047] After obtaining the target's ground speed characteristics and heading change characteristics, the server associates and combines these two characteristics according to the same trajectory point to form a trajectory point motion feature sequence. The trajectory point motion feature is a joint feature used to describe the motion state of a single trajectory point, composed of the target's ground speed characteristics and heading change characteristics; the trajectory point motion feature sequence is a sequence formed by arranging the motion features of multiple trajectory points in the same continuous AIS trajectory sequence according to the trajectory time order. The server will then... The target's ground speed and heading change characteristics are bound to each trajectory point, and the corresponding ship identification, trajectory time, and latitude / longitude position are retained, enabling the motion characteristics of this trajectory point to be written back to the original trajectory point during subsequent state annotation. The expression for the trajectory point's motion characteristics is:

[0048]

[0049] in, Indicates the first The motion characteristics of the trajectory points corresponding to each trajectory point; Indicates the first The target ground speed characteristics corresponding to each trajectory point are obtained by the server through anomaly verification and correction of the initial ground speed characteristics; Indicates the first The heading change characteristics corresponding to each trajectory point are calculated by the server using the degree of heading change between adjacent trajectory points and the corresponding time interval. This expression combines the velocity and steering dimensions into the same trajectory point motion characteristics, so that each trajectory point has both a description of movement speed and a description of steering intensity.

[0050] The server then arranges the motion features of multiple trajectory points in chronological order to form a sequence of trajectory point motion features, the specific expression of which is:

[0051] in, This represents the motion feature sequence of trajectory points corresponding to a continuous AIS trajectory sequence; This represents the motion characteristics of the trajectory point corresponding to the first trajectory point; This indicates the motion characteristics of the trajectory point corresponding to the second trajectory point; Indicates the first The motion characteristics of the trajectory points corresponding to each trajectory point; This represents the total number of trajectory points in a continuous AIS trajectory sequence, obtained by the server by counting the number of valid trajectory points corresponding to the same vessel identifier. This expression preserves the chronological relationship of the fishing vessel's motion characteristics according to the trajectory time, enabling subsequent sliding window processing to extract local continuous motion changes from the trajectory point motion feature sequence.

[0052] The server performs sliding window processing on the trajectory point motion feature sequence based on a preset embedding window length. The preset embedding window length refers to the number of consecutive trajectory point motion features selected when constructing a time-series state vector, used to limit the observation range of a local time period. Sliding window processing involves starting with a trajectory point motion feature, continuously selecting multiple adjacent trajectory point motion features corresponding to the preset embedding window length, and moving the starting point backward after completing the construction of one time-series state vector to generate the next. The server determines the number of time-series state vectors that can be generated based on the length of the trajectory point motion feature sequence and the preset embedding window length; the calculation expression is as follows:

[0053]

[0054] in, This represents the number of time-series state vectors that can be generated. This represents the total number of trajectory point motion features in the trajectory point motion feature sequence, which the server obtains by counting the number of valid trajectory points. This represents the preset embedding window length, which is pre-set by the system or determined based on the clustering quality evaluation results. This expression determines the number of complete windows that can be formed during the sliding window's movement from the start to the end of the sequence by subtracting the preset embedding window length from the total number of trajectory points and adding one, ensuring that each temporal state vector has the same feature length.

[0055] In each sliding window processing step, the server continuously extracts the motion features of multiple adjacent trajectory points from the trajectory point motion feature sequence and concatenates them according to the trajectory time order to form a corresponding temporal state vector. The temporal state vector is a high-dimensional motion state description formed by combining the motion features of multiple trajectory points within a local time period, used to express the speed and heading changes of the fishing vessel within that local time period. The expression for constructing the temporal state vector is:

[0056]

[0057] in, Indicates the first A time-series state vector; Indicates the first Motion characteristics of each trajectory point; Indicates the first Motion characteristics of each trajectory point; Indicates the first Motion characteristics of each trajectory point; Indicates the preset embedded window length; Indicates the starting position of the sliding window, with a value range of [value missing]. This expression concatenates the motion features of multiple consecutive trajectory points into a holistic state description according to the trajectory time sequence, so that each temporal state vector not only contains the velocity and turning information of a single trajectory point, but also the continuous motion change patterns between adjacent trajectory points.

[0058] After generating each temporal state vector, the server establishes a correspondence between the vector and the corresponding vessel identifier, start and end trajectory times, and the covered latitude and longitude range. This allows the server to accurately write back the clustering results to the corresponding AIS trajectory points after subsequent clustering analysis categorizes fishing and navigation states. The resulting set of temporal state vectors retains both the target's speed and heading characteristics relative to the ground, as well as the motion evolution within local timeframes. This ensures that subsequent clustering analysis is no longer based on isolated trajectory points but rather on continuous motion behavior segments to identify the fishing and navigation states of fishing vessels, thereby improving the reliability of fishing vessel activity state identification.

[0059] S130. Perform cluster analysis on the time-series state vector to form fishing state category and navigation state category, and perform state labeling on AIS trajectory data based on fishing state category and navigation state category to form fishing vessel activity trajectory data.

[0060] Specifically, the server first acquires the pre-formed temporal state vectors and synchronously reads the vessel identifier, start trajectory time, end trajectory time, and associated trajectory point positions corresponding to each temporal state vector. The vessel identifier is used to determine which fishing vessel the temporal state vector belongs to; the start trajectory time is used to determine the time of the first trajectory point covered by the temporal state vector; the end trajectory time is used to determine the time of the last trajectory point covered by the temporal state vector; and the associated trajectory point positions are used to record the positions of multiple trajectory points covered by the temporal state vector in the continuous AIS trajectory sequence. Since the temporal state vectors simultaneously contain target-to-ground speed characteristics and heading change characteristics, and their numerical ranges are usually different, if clustering is performed directly, the target-to-ground speed characteristics with larger values ​​tend to dominate distance calculations, weakening the influence of heading change characteristics. Therefore, the server needs to perform feature scaling uniformity processing on the temporal state vectors to ensure that different types of features have comparable contributions in cluster analysis. Feature scaling uniformity processing can adopt a standardization method, the expression of which is:

[0061]

[0062] in, Indicates the first The th time-series state vector Standardized feature values ​​after feature scaling uniformity processing; Indicates the first The th time-series state vector The server reads the original feature values ​​of each feature from the time-series state vector; This represents the timing state vector number, and its value ranges from the first timing state vector to the last timing state vector. This indicates the feature number, which specifically corresponds to the target's ground speed feature or heading change feature at each time position in the time-series state vector; Represents the nth time series state vector in the total time series state vector. The sample mean of each feature is obtained by the server from all... We get the average. Represents the nth time series state vector in the total time series state vector. The sample standard deviation of each feature is determined by the server based on all... Compared to The degree of dispersion is calculated. This expression transforms different features to a similar numerical scale by subtracting the mean and dividing by the standard deviation, thus preventing one feature from suppressing other features due to its large numerical range.

[0063] The server then performs cluster analysis on the time-series state vectors that have undergone feature scale unification processing, based on a preset number of cluster categories. The preset number of cluster categories limits the number of behavioral categories to be formed during cluster analysis; in this scheme, it corresponds to fishing state categories and navigation state categories, and therefore can be set to two. The server first selects two initial category centers from the time-series state vectors that have undergone feature scale unification processing. Each category center is a representative vector of a certain category in the feature space. Then, it calculates the feature distance between each time-series state vector and each category center, and assigns the time-series state vector to the temporary category corresponding to the category center with the smallest feature distance. The expression for calculating the feature distance is:

[0064]

[0065] in, Indicates the first The time-series state vector and the first Feature distance between category centers; Indicates the first The th time-series state vector One standardized feature value; Indicates the first The first category center One central eigenvalue; This represents the category center number, with a value range from the first category center to the category center corresponding to the preset number of cluster categories; This represents the total number of features contained in each temporal state vector, determined by the preset embedding window length and the number of features in the motion features of each trajectory point. This expression measures the closeness between the temporal state vector and the class center in the feature space by accumulating the differences in each feature dimension; the smaller the feature distance, the more similar their motion patterns are.

[0066] After completing the category assignment process, the server performs category center update processing on the temporal state vectors in each temporary category. Category center update processing involves recalculating the average value of all temporal state vectors within the same temporary category across all feature dimensions, and using this average value as the new category center, gradually bringing the category center closer to the concentrated location of similar motion patterns. The expression for category center update is:

[0067]

[0068] in, Indicates the first The category center in the The updated central feature values ​​in each feature dimension; Indicates the first The number of temporal state vectors contained in each temporary category is statistically assigned by the server to the first category. The number of temporal state vectors for each temporary category is obtained; Indicates the first A set of temporal state vectors in each temporary category; Indicates belonging to the first The first temporary category The time-series state vector at the th time series state vector in the th... The standardized feature values ​​are calculated over each feature dimension. This expression averages the temporal state vectors within the same temporary category along the same feature dimension, ensuring that the updated category center represents the overall motion characteristics of that temporary category.

[0069] The server repeatedly executes the class assignment and class center update processes until the changes in class centers satisfy the convergence condition, thus forming multiple stable classes. The convergence condition can be that the change in class centers between two consecutive iterations is less than a preset convergence threshold, or that the number of iterations reaches a preset maximum number of iterations. The expression for calculating the change in class centers is:

[0070]

[0071] in, Indicates the first After the iteration, relative to the first Change in class center in each iteration; This indicates the number of preset clustering categories, which in this scheme correspond to the fishing status category and the navigation status category; This represents the total number of features contained in each category center; Indicates the first After the nth iteration The category center in the Central feature values ​​in each feature dimension; Indicates the first After the nth iteration The category center in the The expression measures the central feature value across each feature dimension. This expression measures whether the class center still shifts significantly between adjacent iterations. When the change in the class center is less than a preset convergence threshold, it indicates that the class division has stabilized, and the server stops iterating and outputs a stable class.

[0072] After forming multiple stable categories, the server determines the fishing state category and the navigation state category based on the ground speed characteristic distribution and heading change characteristic distribution corresponding to each stable category. The ground speed characteristic distribution reflects the overall level of the fishing vessel's movement speed within the same stable category, while the heading change characteristic distribution reflects the overall level of the fishing vessel's turning intensity within the same stable category. The server restores the temporal state vectors within the same stable category to the corresponding target ground speed characteristic and heading change characteristic, and calculates their speed center value and turning center value respectively. When a stable category exhibits a low speed center value and a high turning center value, the server identifies this stable category as the fishing state category, because fishing operations typically require low-speed towing, net setting, net retrieval, or detours, accompanied by frequent directional adjustments. When another stable category exhibits a high speed center value and a low turning center value, the server identifies this stable category as the navigation state category, because transoceanic navigation typically exhibits higher speeds and more stable headings. The expressions for the speed center value and turning center value are:

[0073]

[0074]

[0075] in, Indicates the first The velocity center value corresponding to each stable category; Indicates the first The turning center value corresponding to each stable category; Indicates the first The number of temporal state vectors in each stable category; Indicates the preset embedded window length; Indicates the first The th time-series state vector The target ground speed features corresponding to the motion features of each trajectory point are read by the server from the correlation between the temporal state vector and the motion features of the trajectory points; Indicates the first The th time-series state vector The heading change features corresponding to the motion features of each trajectory point are read by the server from the correlation between the temporal state vector and the motion features of the trajectory points; This represents the position number of the trajectory point motion feature in the time-series state vector, with values ​​ranging from the first window position to the [missing information]. The expression above extracts the typical motion state of a stability category by averaging the velocity and steering characteristics of all window positions within the same stability category.

[0076] The server generates status labels based on fishing status and navigation status categories, and writes these status labels back to the corresponding trajectory points in the AIS trajectory data to form the initial status labeling result. Status labels are behavioral markers attached to trajectory points, used to describe the fishing vessel's activity status corresponding to that trajectory point. The initial status labeling result is the first version of the status recognition result formed after mapping the category semantics obtained from clustering back to the AIS trajectory data. Since each temporal state vector covers multiple consecutive trajectory points, and the same trajectory point may be covered by multiple sliding windows, when writing back the status labels, the server first assigns the category label of each temporal state vector to the associated trajectory point positions it covers. Then, it counts the multiple category labels received for the same trajectory point and uses the category label that appears most frequently as the initial status label for that trajectory point. The expression for determining the initial status label is:

[0077]

[0078] in, Indicates the first Initial state labels for each trajectory point; Indicates the category label to be compared. Indicates the fishing status label. Indicates navigation status label; Indicates the first The set of associated trajectory point locations covered by each temporal state vector is recorded by the server when constructing the temporal state vector; Indicates the first The category labels corresponding to each temporal state vector are obtained by cluster analysis and behavioral semantic determination. This indicates an indicator function that takes the value 1 when the condition inside the parentheses is true, and 0 otherwise. This expression selects the category label with the highest statistical frequency. By statistically analyzing the category labels given by multiple temporal state vectors covering the same trajectory point, it determines the most stable initial state label for that trajectory point, thereby reducing the impact of misjudgments in a single window on trajectory point labeling.

[0079] The server performs a state consistency check on the initial state labeling results and corrects isolated state segments based on the state labels corresponding to adjacent trajectory points to form the final state labeling results. The state consistency check verifies whether the state labels in the same continuous AIS trajectory sequence conform to the temporal continuity of fishing vessel behavior. An isolated state segment refers to a short segment with a short duration, a small number of trajectory points, and sandwiched between identical state labels; for example, a very short fishing state segment appearing in the middle of a continuous navigation state, or a very short navigation state segment appearing in the middle of a continuous fishing state. The server determines whether correction is needed based on the duration of the isolated state segment, the number of trajectory points, and the adjacent state labels before and after it. If the duration of an isolated state segment is less than a preset state duration threshold, and its adjacent state labels are the same, the isolated state segment is corrected to match the adjacent state labels, thus forming the final state labeling results. The expression for calculating the duration of an isolated state segment is:

[0080]

[0081] in, Indicates the first The duration of an isolated state segment; Indicates the first The trajectory time of the first trajectory point in an isolated state segment; Indicates the first The trajectory time of the last trajectory point in an isolated state segment; Indicates the first The starting trajectory point number of an isolated state segment; Indicates the first The end trajectory point number of an isolated state segment. This expression is used to determine whether an isolated state segment is shorter than the reasonable behavior duration. If it is shorter than a preset state duration threshold, it indicates that the segment is more likely caused by trajectory noise or cluster boundary fluctuations, and state correction should be performed.

[0082] The server generates fishing vessel activity trajectory data based on the final status annotation results. This data is behavioral trajectory data formed by adding motion features, status labels, and trajectory interruption markers to the target AIS trajectory data. It includes vessel identification, trajectory time, latitude and longitude position, ground speed characteristics, heading change characteristics, status labels, and trajectory interruption markers. Vessel identification distinguishes different fishing vessels; trajectory time indicates the temporal sequence of trajectory points; latitude and longitude position indicates the spatial location of trajectory points; ground speed characteristics indicate the speed of the fishing vessel's movement; heading change characteristics indicate the intensity of the fishing vessel's turning; status labels distinguish between fishing and navigation states; and trajectory interruption markers record intervals where the AIS trajectory is missing for extended periods. This fishing vessel activity trajectory data serves both as a basis for extracting trajectory points corresponding to subsequent fishing state categories and as a basis for spatiotemporal cross-matching with VIIRS / DNB nighttime light remote sensing data, thereby supporting the identification of unmatched nighttime light targets and concealed nighttime operating targets.

[0083] S140. Perform spatiotemporal cross-matching processing on VIIRS / DNB nighttime light remote sensing data and fishing vessel activity trajectory data to identify unmatched nighttime light targets, and identify concealed nighttime operation targets based on the radiance distribution of unmatched nighttime light targets, so as to form abnormal fishing vessel activity identification results.

[0084] Specifically, the server first performs quality screening on the VIIRS / DNB nighttime light remote sensing data. VIIRS / DNB nighttime light remote sensing data is nighttime sea surface light detection data acquired by a visible infrared imaging radiometer in the day and night bands. Nighttime light detection points are spatial observation points in the VIIRS / DNB nighttime light remote sensing data that can characterize luminous targets on the sea surface at night. Each nighttime light detection point includes at least the nighttime detection time, nighttime light detection location, radiance value, detection confidence level, and quality label. The nighttime detection time characterizes the time when the nighttime light detection point was observed by the satellite; the nighttime light detection location characterizes the latitude and longitude of the nighttime light detection point in the target sea area; the radiance value characterizes the luminous intensity of the nighttime light detection point; the detection confidence level characterizes the degree to which the nighttime light detection point belongs to a real sea surface light target; and the quality label characterizes whether the nighttime light detection point is affected by cloud cover, moonlight interference, shoreline light spillover, fixed marine facility lighting, or sensor malfunctions. The server first removes nighttime light detectors with abnormal quality indicators, then removes nighttime light detectors whose detection confidence level does not reach a preset confidence threshold, and retains nighttime light detectors located in the target sea area and target time period to form candidate nighttime light targets. Candidate nighttime light targets are not final abnormal targets, but rather valid nighttime light targets that have undergone quality screening and can be used for subsequent spatiotemporal cross-matching.

[0085] The selection criteria for candidate luminous targets can be expressed by a target retention function as follows:

[0086] in, Indicates the first The markers for each night-light detection point are retained when... At that time, the server will... Several night-light detection points are retained as candidate night-light targets when At that time, the server removed the first One night-light detection point; Indicates the first The quality identifier of each night light detection point is read by the server from VIIRS / DNB night light remote sensing data; This indicates a preset valid quality indicator, used to indicate a quality status that is not affected by cloud cover, moonlight interference, shoreline light spillover, fixed offshore facility lighting, or sensor malfunctions. Indicates the first The detection confidence level of each night light detection point is read by the server from VIIRS / DNB night light remote sensing data; This indicates a preset confidence threshold, which is set by the system based on the product quality level or historical false detection rate of the VIIRS / DNB nighttime light remote sensing data. Indicates the first The night light detection locations of each night light detection point are read by the server from VIIRS / DNB night light remote sensing data; It indicates the spatial extent corresponding to the target sea area, which is determined by the boundary of the target sea area; Indicates the first The night light detection time for each night light detection point is read by the server from VIIRS / DNB night light remote sensing data; The target time period is determined by the start and end times. This indicates an indicator function that takes the value 1 when the condition within the parentheses is true, and 0 otherwise. This function uses a combination of quality status, confidence level, spatial range, and temporal range to ensure that candidate night-light targets entering subsequent processing possess valid observational attributes.

[0087] The server then constructs a trajectory matching index based on the fishing vessel activity trajectory data. This data is behavioral trajectory data formed after prior status annotation of the AIS trajectory data, and includes at least the vessel identifier, trajectory time, latitude and longitude position, ground speed characteristics, heading change characteristics, status labels, and trajectory interruption markers. The trajectory matching index is a data retrieval structure that organizes the fishing vessel activity trajectory data according to trajectory time and latitude and longitude position, used to quickly find whether a corresponding AIS trajectory point exists around a candidate nighttime luminous target. The server aggregates the fishing vessel activity trajectory data according to a unified time unit and a unified spatial unit, recording trajectory points within the same time unit and spatial unit in the same index position, and retaining status labels and trajectory interruption markers. A trajectory interruption marker is a marker formed when there is a long period of absence between adjacent AIS trajectory points corresponding to the same vessel identifier, used to characterize the possible AIS silence. The server determines the trajectory silence interval based on the trajectory interruption marker. The trajectory silence interval refers to the time range within which the same vessel identifier lacks AIS trajectory records between the previous and subsequent valid trajectory points, and combines this with the spatial positions of the preceding and subsequent valid trajectory points to determine the corresponding spatiotemporal neighborhood.

[0088] The silent interval of a trajectory can be determined by the time interval between adjacent trajectory points, expressed as:

[0089] in, Indicates the first The trajectory point and the first A trajectory interruption marker between trajectory points, when At that time, the server determines that there is a silent interval between adjacent trajectory points. At that time, the server determines that there is no silent interval between adjacent trajectory points; Indicates the first number corresponding to the same ship identification The server reads the trajectory time of each trajectory point from the fishing boat activity trajectory data; Indicates the first number corresponding to the same ship identification The server reads the trajectory time of each trajectory point from the fishing boat activity trajectory data; This indicates the preset silence time threshold, which is set by the system based on the normal broadcast interval of AIS, the range of reception errors, and the sensitivity of the monitoring scenario to silence behavior; This indicates an indicator function that takes a value of 1 when the time interval between adjacent trajectory points is greater than a preset silence time threshold, and a value of 0 otherwise. This expression identifies possible AIS silence by identifying time gaps between adjacent AIS trajectory points, providing a basis for subsequent determination of whether unmatched nighttime targets are related to trajectory silence.

[0090] The server performs spatiotemporal cross-matching processing on candidate night-light targets and their trajectory matching indexes. This process determines whether a candidate night-light target can be found in the fishing vessel activity trajectory data based on both temporal and spatial proximity. For each candidate night-light target, the server searches the trajectory matching index for trajectory points that are adjacent in both time and location of night-light detection. When a trajectory point satisfying both temporal and spatial proximity conditions exists, the server identifies the corresponding candidate night-light target as a matched night-light target and records its corresponding vessel identifier, trajectory time, latitude and longitude, and status label. When no trajectory point satisfying both conditions exists, the server identifies the corresponding candidate night-light target as an unmatched night-light target and retains its night-light detection time, location, radiance value, and detection confidence level. A matched night-light target indicates that the night-light target can be supported by an AIS trajectory; an unmatched night-light target indicates that the night-light target lacks AIS trajectory support, potentially corresponding to AIS being off, AIS signal missing, or a non-fishing vessel background light source.

[0091] The decision expression for spatiotemporal cross-matching is:

[0092] in, Indicates the first Matching tags for candidate luminous targets, when At that time, the server will... One candidate luminous target was identified as the matching luminous target when At that time, the server will... The candidate luminous targets were identified as unmatched luminous targets; This represents the set of searchable trajectory points in the trajectory matching index, which is constructed from fishing vessel activity trajectory data; Indicates the first The server reads the night-light detection time of each candidate night-light luminous target from the candidate night-light luminous targets; Indicates the first The server reads the trajectory time of each trajectory point from the trajectory matching index; This indicates the preset matching time threshold, which is set by the temporal resolution of VIIRS / DNB nighttime light remote sensing data, the AIS trajectory sampling interval, and the matching tolerance. Indicates the first The server reads the night-light detection locations of the candidate night-light targets. Indicates the first The server reads the latitude and longitude positions of each trajectory point from the trajectory matching index; Indicates the first The candidate luminescent targets and the first The spatial distance between each trajectory point is calculated from their latitude and longitude positions; This indicates the preset matching spatial threshold, which is set by the spatial resolution of the VIIRS / DNB nighttime light remote sensing data, the AIS positioning error, and the target sea area positioning error. This means that if a trajectory point satisfies both the temporal and spatial proximity conditions, the corresponding candidate night-light target is identified as a matching night-light target. This expression uses a combination of temporal and spatial thresholds to avoid incorrect matches based solely on spatial proximity or temporal proximity.

[0093] The server performs a silent association determination based on the spatiotemporal relationship between unmatched night-light targets and the silent interval of the trajectory. This silent association determination identifies whether an unmatched night-light target has a spatiotemporal correspondence with the silent AIS trajectory. The server reads the night-light detection time and location of each unmatched night-light target and retrieves the start and end times of the silent trajectory interval and its corresponding spatial neighborhood. When the night-light detection time of an unmatched night-light target falls within the start and end time range of a certain silent trajectory interval, and the night-light detection location of the unmatched night-light target is located within the corresponding spatial neighborhood of that silent trajectory interval, the server identifies the unmatched night-light target as a silently associated night-light target. When an unmatched night-light target does not meet the above conditions, the server identifies it as an independent unmatched night-light target. A silently associated night-light target indicates that the target not only lacks AIS trajectory support but also has a spatiotemporal association with the AIS trajectory interruption, thus making it more likely to correspond to covert operations under AIS silent conditions. Independent unmatched night-light targets require further judgment based on radiance and spatial concentration.

[0094] The expression for silent association determination is:

[0095] in, Indicates the first The silent association state of an unmatched luminous target, when At that time, the server will... Unmatched luminous targets were identified as silently associated luminous targets. At that time, the server will... The unmatched luminous targets were identified as independent unmatched luminous targets; This represents the set of silent intervals of the trajectory, determined by the trajectory interruption marker; Represents any silent interval of a trajectory within the set of silent intervals for trajectories; Indicates the first The start time of each trajectory silent interval is determined by the trajectory time of the last valid trajectory point before the trajectory is interrupted; Indicates the first The end time of each trajectory silent interval is determined by the trajectory time of the first valid trajectory point after trajectory recovery; Indicates the first Night light detection time for one unmatched night light target; Indicates the first The night light detection locations of the unmatched night light targets; Indicates the first The spatial neighborhood corresponding to each trajectory silent interval is determined by the position of the last valid trajectory point before the trajectory is interrupted, the position of the first valid trajectory point after the trajectory is restored, and the reasonable navigation range. Indicates the first The unmatched luminous target and the first The minimum spatial distance between spatial neighbors corresponding to each silent interval of a trajectory; This represents the preset silent correlation spatial threshold, set by AIS positioning error, night light detection spatial resolution, and the possible movement range during the trajectory silence period. This expression also constrains the time during which no matching night light target falls within the trajectory silence interval, and the spatial location is close to the corresponding spatial neighborhood of the trajectory silence interval, thereby avoiding misclassification of irrelevant night light targets as AIS silent related targets.

[0096] The server constructs a radiance distribution based on the radiance values ​​of unmatched night-light targets. Radiance values ​​are observed values ​​characterizing the luminous intensity of night-light targets. Light-attracting fishing operations typically require the use of high-intensity fish-attracting lights, therefore their radiance values ​​are usually higher than those of weak background noise, ordinary navigation lights, or low-intensity sea surface light sources. The server collects the radiance values ​​of all unmatched night-light targets and performs density estimation or histogram statistics on the radiance values ​​to form a radiance distribution. When the radiance distribution exhibits a bimodal structure with both low and high brightness peaks, the server determines the local trough between the low and high brightness peaks as the brightness segmentation threshold. The brightness segmentation threshold is used to classify unmatched night-light targets into low-brightness unmatched targets and high-brightness unmatched targets. Low-brightness unmatched targets are more likely to correspond to weak background light sources or ordinary navigation lights, while high-brightness unmatched targets are more likely to correspond to targets used in high-intensity light operations.

[0097] The radiance distribution can be expressed by kernel density estimation as:

[0098] in, Indicates the radiance value The estimated radiance distribution density at that location; This indicates the number of unmatched luminous targets, which the server obtains by counting the set of unmatched luminous targets. The kernel density estimation bandwidth is determined by the system based on the dispersion of the radiance values ​​or a preset smoothness level. This represents the kernel function, used to measure the radiance value of each unmatched night-light target relative to the current location. The degree of impact; Indicates the first The server reads the radiance value of each unmatched luminous target from the unmatched luminous targets. This represents the radiance value to be evaluated. This expression expands the radiance value of each unmatched night-light target into a smooth contribution and then superimposes all contributions to obtain the overall radiance distribution of the unmatched night-light target, thus facilitating the identification of low-brightness peaks, high-brightness peaks, and the brightness segmentation positions between them.

[0099] The brightness segmentation threshold can be determined by the local minimum density location between the two peaks, as expressed by:

[0100] in, Indicates the brightness segmentation threshold; This indicates the radiance position corresponding to the low brightness peak, which is obtained by the server identifying the low brightness peak in the radiance distribution; This indicates the radiance position corresponding to the high-brightness peak, which is obtained by the server identifying the high-brightness peak in the radiance distribution; This indicates the radiance search interval between the low-brightness peak and the high-brightness peak. This expression represents the radiance distribution density. By finding the location of the lowest distribution density between the low-brightness peak and the high-brightness peak, it separates two brightness groups, providing an objective basis for classifying low-brightness mismatched targets and high-brightness mismatched targets.

[0101] The server classifies unmatched night-light targets into low-brightness unmatched targets and high-brightness unmatched targets based on a brightness segmentation threshold. Specifically, when the radiance value of an unmatched night-light target is less than the brightness segmentation threshold, the server classifies it as a low-brightness unmatched target; when the radiance value of an unmatched night-light target is greater than or equal to the brightness segmentation threshold, the server classifies it as a high-brightness unmatched target. The segmentation expression is:

[0102]

[0103] in, Indicates the first A high-brightness marker that does not match a night-light target, when At that time, the server will... One unmatched luminous target was identified as a high-brightness unmatched target. At that time, the server will... The unmatched night-light targets were identified as low-brightness unmatched targets. Indicates the first The server reads the radiance value of each unmatched luminous target from the unmatched luminous targets. The brightness segmentation threshold is determined by the local minimum density location between the low brightness peak and the high brightness peak in the radiance distribution. This indicates an indicator function. This expression allows the brightness attribute of unmatched nighttime luminous targets to be transformed into a clear target category attribute, providing a judgment condition for target identification in subsequent nighttime covert operations.

[0104] The server identifies covert nighttime operation targets based on high-brightness unmatched targets and silently associated nighttime luminous targets. Covert nighttime operation targets refer to potential fishing vessel operation targets that lack AIS trajectory support, possess high-intensity nighttime radiation characteristics, and are silently associated with AIS trajectories. For each unmatched nighttime luminous target, the server simultaneously checks the high-brightness marker and silent association status. When an unmatched nighttime luminous target is both a high-brightness unmatched target and a silent associated nighttime luminous target, the server identifies it as a covert nighttime operation target. When an unmatched nighttime luminous target is a high-brightness unmatched target but not a silent associated nighttime luminous target, the server can identify it as a suspected covert nighttime operation target and differentiate it within the target type. When an unmatched nighttime luminous target is a low-brightness unmatched target, the server does not identify it as a covert nighttime operation target. The identification expression for covert nighttime operation targets is:

[0105]

[0106] in, Indicates the first A concealed operation marker that did not match a luminous target, when At that time, the server will... The unmatched luminous targets were identified as targets for covert nighttime operations. At that time, the server will not send the first The unmatched luminous targets were identified as targets for covert nighttime operations. Indicates the first High-brightness markers of unmatched night-light targets are obtained by dividing them using brightness segmentation thresholds; Indicates the first The silent association state of an unmatched night-light target is determined by the spatiotemporal relationship between the unmatched night-light target and the silent interval of the trajectory. This expression avoids mistaking a normal non-AIS night-light background for a concealed operation target by simultaneously satisfying the high brightness condition and the silent association condition, and also avoids misjudgment when relying solely on AIS silence without strong light operation evidence.

[0107] The server ultimately generates abnormal fishing vessel activity identification results based on the concealed nighttime fishing targets. These results are generated from structured recordings of concealed nighttime fishing targets, including at least nighttime light detection time, nighttime light detection location, radiance value, silent association status, and target type. Nighttime light detection time indicates the time of appearance of the concealed nighttime fishing target; nighttime light detection location indicates the spatial location of the concealed nighttime fishing target; radiance value indicates the intensity of the concealed nighttime fishing target's operating lights; silent association status indicates whether the concealed nighttime fishing target has a spatiotemporal association with a silent trajectory interval; and target type distinguishes between concealed nighttime fishing targets, suspected concealed nighttime fishing targets, and non-concealed fishing targets. The resulting abnormal fishing vessel activity identification results can not only be used by the regulatory side to identify potential concealed fishing activities under AIS silent conditions, but also serve as supplementary activity evidence for missing AIS trajectory areas during the subsequent construction of the fishing vessel activity intensity field, creating a continuous technical correlation between the fishing vessel activity identification results and the spatiotemporal prediction process.

[0108] S150. Based on the identification results of abnormal fishing vessel activities and the fishing vessel activity trajectory data corresponding to the fishing status category, a fishing vessel activity intensity field is constructed, and combined with marine environmental data to form a continuous spatiotemporal sequence sample.

[0109] Specifically, the server first establishes a unified spatial grid and a unified time step covering the target sea area. The unified spatial grid consists of multiple spatial units divided by fixed longitude and latitude intervals, used to carry fishing vessel activity trajectory data, abnormal fishing vessel activity identification results, and marine environmental data. The unified time step consists of multiple time units divided by fixed time intervals for the target time period, used to ensure that data from different sampling frequencies can be included at the same time scale. The server determines the extent of each spatial unit based on the longitude and latitude boundaries of the target sea area, and the extent of each time unit based on the start and end times of the target time period. Then, it maps each fishing status trajectory point in the fishing vessel activity trajectory data corresponding to the fishing status category to the corresponding spatial unit and time unit. A fishing status trajectory point refers to an AIS trajectory point with a status label of fishing status category, which can represent the actual fishing activity under AIS visibility conditions. The expressions for determining the spatial unit index and the time unit index are:

[0110]

[0111]

[0112] in, Indicates the first Spatial unit index corresponding to each fishing status trajectory point; Indicates the first The server reads the longitude of each fishing status trajectory point from the fishing vessel activity trajectory data corresponding to the fishing status category; Indicates the first The server reads the latitude of each fishing status trajectory point from the fishing vessel activity trajectory data corresponding to the fishing status category; The minimum longitude of the target sea area is determined by the boundary of the target sea area; The minimum latitude representing the target sea area is determined by the boundary of the target sea area; This indicates the grid spacing of the unified spatial grid in the longitude direction, which is set by the system according to the prediction resolution requirements; This indicates the grid spacing of the unified spatial grid in the latitudinal direction, which is set by the system according to the prediction resolution requirements; This indicates rounding down, used to convert continuous latitude and longitude positions into spatial unit numbers; Indicates the first The time unit index corresponding to each fishing status trajectory point; Indicates the first The server reads the trajectory time of each fishing status trajectory point from the fishing vessel activity trajectory data corresponding to the fishing status category; Indicates the start time of the target time period; This represents a uniform time step, set by the system based on the predicted time resolution. The above expression determines the spatial and temporal units to which the fishing status trajectory points belong by using the offset of spatial location relative to the boundary of the target sea area and the offset of trajectory time relative to the start time of the target time period.

[0113] After completing the spatial and temporal mapping processing of fishing status trajectory points, the server counts the number of fishing status trajectory points in each spatial and temporal unit. It then merges duplicate trajectory points of the same fishing vessel within the same time unit, based on vessel identifiers, to form the AIS fishing activity intensity. AIS fishing activity intensity refers to the strength of fishing activities statistically derived from visible AIS trajectory data, characterizing the density of fishing activities recorded by AIS trajectories within a specific spatial and temporal unit. To avoid over-counting due to a large number of trajectory points for the same fishing vessel in high-frequency AIS reporting scenarios, the server can first deduplicate based on vessel identifiers before counting the number of fishing vessels or their trajectory contributions within the corresponding spatial and temporal units. The expression for calculating AIS fishing activity intensity is:

[0114]

[0115] in, Indicates the first The spatial unit and the first The intensity of AIS fishing activity corresponding to each time unit; This represents the total number of trajectory points in the fishing vessel activity trajectory data, which the server obtains by counting the fishing vessel activity trajectory data. Indicates the first Spatial unit index corresponding to each trajectory point; Indicates the first The time unit index corresponding to each trajectory point; Indicates the first The status labels of each trajectory point are read by the server from the fishing vessel activity trajectory data; Indicates the fishing status label; Indicates the first The trajectory contribution coefficient of each trajectory point is 1 when counting by trajectory points, and when counting by deduplication based on the same ship identifier, it is the reciprocal of the number of trajectory points of the same ship identifier in the same spatial unit and the same time unit. This indicates an indicator function, which takes a value of 1 when the condition within the parentheses is true, and a value of 0 otherwise. This expression filters trajectory points within the same spatial and temporal unit that are labeled as "fishing," and accumulates their trajectory contributions to form the intensity of fishing activity under AIS visibility conditions.

[0116] The server then performs spatial and temporal mapping processing on the abnormal fishing vessel activity identification results based on a unified spatial grid and a unified time step. The abnormal fishing vessel activity identification results are the covert operation identification results formed by the spatiotemporal cross-matching of VIIRS / DNB nighttime light remote sensing data and fishing vessel activity trajectory data. These results include at least the nighttime light detection time, nighttime light detection location, radiance value, silent association state, and target type. The nighttime light detection location is used to determine the spatial cell into which the nighttime covert operation target falls; the nighttime light detection time is used to determine the time cell into which the nighttime covert operation target falls; the radiance value is used to characterize the intensity of the nighttime light operation; the silent association state is used to characterize the degree of association between the nighttime covert operation target and the silent interval of the trajectory; and the target type is used to distinguish between nighttime covert operation targets and suspected nighttime covert operation targets. The server maps each abnormal fishing vessel activity identification result to a corresponding spatial cell and a corresponding time cell, and determines the intensity of the covert operation activity based on the radiance value, silent association state, and target type. The expression for calculating the intensity of the covert operation activity is:

[0117]

[0118] in, Indicates the first The spatial unit and the first The intensity of covert operations corresponding to each time unit; This indicates the number of luminous targets in the abnormal fishing vessel activity identification results, which the server obtains by statistically analyzing the abnormal fishing vessel activity results. Indicates the first The spatial cell index corresponding to each night-light-emitting target is obtained by mapping the night-light-emitting detection location through a unified spatial grid. Indicates the first The time unit index corresponding to each night-light-emitting target is obtained by mapping the night-light-emitting detection time to a uniform time step. Indicates the first The server reads the radiance value of each luminous target from the results of abnormal fishing vessel activity identification. This represents the radiance normalization function, used to convert radiance values ​​into luminance contributions that can participate in the accumulation of activity intensity. Indicates the first The silent association state of a luminous target is set to 1 when the luminous target has a spatiotemporal association with the silent interval of the trajectory, and 0 otherwise. This represents the silent correlation enhancement coefficient, used to increase the contribution of nighttime luminous targets associated with the silent interval of the trajectory to the intensity of covert operations. Indicates the first The target type coefficient for each luminous target is set to a higher value when the target type is a nighttime covert operation target and a lower value when the target type is a suspected nighttime covert operation target. This represents the indicator function. The expression reflects the intensity of light operations through radiance values, the strength of silent AIS evidence through silent association states, and the recognition confidence level through target type, thus enabling the intensity of covert operations to reflect the potential contribution of fishing activities under conditions where AIS is not visible.

[0119] The radiance normalization function can be expressed as follows:

[0120] in, Indicates the first Normalized values ​​of the radiance of a luminous target; Indicates the first The server reads the radiance value of each luminous target from the results of abnormal fishing vessel activity identification. This represents the minimum radiance value of all luminous targets in the abnormal fishing vessel activity identification results. The server calculates this by statistically analyzing all... get; This represents the maximum radiance value of all luminous targets in the abnormal fishing vessel activity identification results. The server calculates this by statistically analyzing all... get; This expression, which prevents the denominator from being an extremely small positive number (zero), is preset by the system. By converting radiance values ​​to a relative intensity range, this expression allows the brightness contribution of different nighttime luminous targets to participate in the intensity calculation of covert operations at the same scale.

[0121] The server performs a fusion process on the AIS fishing activity intensity and the covert operation activity intensity to form a fishing vessel activity intensity field. This field is a spatiotemporal matrix with a unified spatial grid and a unified time step, used to characterize the spatial and temporal distribution of fishing activities within the target sea area. The server uses the AIS fishing activity intensity as the contribution of visible fishing activities and the covert operation activity intensity as the contribution of silent fishing activities, determining the correlation between the two based on trajectory interruption markers and silent association states. When the nighttime luminous target corresponding to the covert operation activity intensity has a silent association with the trajectory silent interval indicated by the trajectory interruption marker, the server uses this covert operation activity intensity as a missing supplement to the AIS fishing activity intensity; when the covert operation activity intensity does not establish a association with a specific trajectory interruption marker but the target type is still a suspected nighttime covert operation target, the server treats it as an independent silent fishing activity contribution and assigns it a lower fusion contribution. The fusion expression for the fishing vessel activity intensity field is:

[0122]

[0123] in, Indicates the first The spatial unit and the first The numerical values ​​of the fishing vessel activity intensity field corresponding to each time unit; Indicates the first The spatial unit and the first The intensity of AIS fishing activity corresponding to each time unit; Indicates the first The spatial unit and the first The intensity of covert operations corresponding to each time unit; Indicates the first The spatial unit and the first The concealed operation fusion coefficient corresponding to each time unit is determined by the silent association state, target type, and trajectory interruption marker matching of the nighttime targets within that spatial and time unit. This expression is based on AIS visible fishing activities and incorporates concealed operation activities as a supplement, enabling the fishing vessel activity intensity field to simultaneously represent both AIS trajectory visible operations and AIS trajectory silent operations.

[0124] The concealed operation fusion coefficient can be determined by the following expression:

[0125] in, Indicates the first The spatial unit and the first The concealed operation fusion coefficient corresponding to each time unit; The base fusion coefficient represents the basic contribution of a suspected nighttime covert operation target in the absence of a clear trajectory silent association. This represents the silent correlation adjustment coefficient, used to indicate the enhancing effect of the silent correlation state on the fusion strength; Indicates the first The spatial unit and the first The average silent association state of a night-light target within a time unit is obtained by averaging the silent association states of all night-light targets within that spatial unit and that time unit. This represents the trajectory interruption adjustment coefficient, used to indicate the enhancement effect of trajectory interruption markers on fusion strength; Indicates the first The spatial unit and the first The trajectory interruption association status within a time unit is determined by whether there is a trajectory interruption marker in that spatial and time unit that is spatiotemporally adjacent to the night-light target. The value is 1 if it exists and 0 if it does not exist. This expression determines the contribution ratio of the intensity of covert operation activities to the fusion by combining the basic contribution, silent association contribution, and trajectory interruption contribution, thereby avoiding treating all unmatched night-light targets equally.

[0126] The server performs temporal and spatial continuity checks on the fishing vessel activity intensity field, and corrects for abnormal activity intensities based on the distribution of fishing vessel activity intensities corresponding to adjacent time and spatial units. Temporal continuity checks determine whether there are abnormal spikes or drops in fishing vessel activity intensity within adjacent time units that do not conform to the activity evolution pattern. Spatial continuity checks determine whether there are isolated abnormally high or low values ​​in fishing vessel activity intensity between adjacent spatial units within the same time unit. The server compares the fishing vessel activity intensity field value for each spatial unit and each time unit with the activity intensities of its adjacent time and spatial units. When the difference exceeds a preset continuity threshold and there is a lack of corresponding AIS fishing activity intensity or covert operation activity intensity support, the corresponding value is identified as an abnormal activity intensity, and corrected using the activity intensities of adjacent spatial and time units to form the target fishing vessel activity intensity field. The verification expression for abnormal activity intensity is:

[0127]

[0128] in, Indicates the first The spatial unit and the first The abnormal activity intensity flag corresponding to each time unit, when At that time, the server determined that there was abnormal activity intensity at that location. At that time, the server determined that there was no abnormal activity intensity at that location; This represents the numerical value of the fishing vessel activity intensity field before correction; Indicates the first The spatial unit and the first The reference value of the neighborhood activity intensity of each time unit is obtained by statistical analysis of the activity intensity field of fishing boats in adjacent spatial units and adjacent time units. This indicates a preset continuity threshold, determined by the historical fluctuation range of activity intensity. Indicates the intensity of AIS fishing activities; Indicates the intensity of covert operations; This represents the threshold for evidence of activity, used to determine whether the corresponding spatial and temporal units lack sufficient evidence of activity. This indicates the indicator function. This expression only identifies fishing vessel activity intensity as abnormal when the value deviates from the neighborhood reference value and lacks supporting evidence from AIS or nighttime light, thus avoiding misclassification of the true high-intensity operation area.

[0129] The corrected expression for the intensity of abnormal activity is:

[0130] in, Indicates the first The spatial unit and the first The numerical values ​​of the target fishing vessel activity intensity field corresponding to each time unit; Indicates the intensity of abnormal activity; This represents the numerical value of the fishing vessel activity intensity field before correction; This represents a reference value for the intensity of activity in the neighborhood. This expression retains the original value when no abnormal activity intensity exists, and replaces it with the reference value when abnormal activity intensity exists, making the activity intensity field of the target fishing vessel more continuous and stable in time and space.

[0131] The server performs spatial and temporal alignment processing on marine environmental data based on a unified spatial grid and a unified time step to generate wind field, ocean current, and chlorophyll concentration states corresponding to the target fishing vessel activity intensity field. Marine environmental data includes wind field data, ocean current data, and chlorophyll concentration data; wind field state refers to wind speed, wind direction, or wind field components under a unified spatial grid and time step; ocean current state refers to current velocity, current direction, or ocean current components under a unified spatial grid and time step; chlorophyll concentration state refers to the chlorophyll concentration distribution under a unified spatial grid and time step. Since marine environmental data often comes from different sources and their spatial and temporal resolutions are not entirely consistent, the server needs to map each type of marine environmental data to the same spatial and temporal units as the target fishing vessel activity intensity field. When the marine environmental data resolution is higher than the unified spatial grid, the server averages multiple environmental pixels falling within the same spatial unit; when the marine environmental data resolution is lower than the unified spatial grid, the server interpolates based on adjacent environmental pixels; when the marine environmental data temporal resolution is inconsistent with the unified time step, the server performs temporal alignment based on adjacent time records. The expression for mapping high-resolution environmental data to a unified spatial grid is:

[0132]

[0133] in, Indicates the first The spatial unit and the first The first time unit corresponding to the first Class environment state; This represents environmental data types, which can correspond to wind field status, ocean current status, or chlorophyll concentration status. Indicates falling into the first The spatial unit and the first The first time unit A set of environment pixels, obtained by the server through mapping environment pixel locations to environment recording times; Indicates the number of environmental pixels in the environmental pixel set; Indicates the first The first environmental pixel corresponds to the first The environmental numerical values ​​are read by the server from marine environmental data. This expression transforms the marine environmental data into an environmental state consistent with the activity intensity field of the target fishing vessel by averaging environmental pixels within the same spatial and temporal unit.

[0134] The server combines the target fishing vessel activity intensity field, wind field status, ocean current status, and chlorophyll concentration status into a single frame of multi-source activity environment data. Channel combination refers to arranging different types of data according to characteristic channels within the same spatial and temporal unit, so that each spatial unit simultaneously contains fishing vessel activity intensity, wind field status, ocean current status, and chlorophyll concentration status. Single-frame multi-source activity environment data refers to multi-channel spatial data corresponding to a single temporal unit, which includes both fishing vessel activity information and marine environmental information. The expression for channel combination is:

[0135]

[0136] in, Indicates the first Within the [number] time unit, the [number]th Single-frame multi-source activity environment data corresponding to each spatial unit; This represents the numerical value of the target fishing vessel's activity intensity field; The server indicates the wind field status, which is obtained from the wind field data through spatial and temporal alignment processing. The representation of ocean current status is obtained by the server from ocean current data through spatial and temporal alignment processing. This expression represents the chlorophyll concentration state, obtained by the server from chlorophyll concentration data through spatial and temporal alignment processing. This expression unifies the activity intensity and environmental state within the same spatial unit into multi-channel features, providing an input structure for subsequent model learning of the coupling relationship between activities and the environment.

[0137] The server performs continuous sequence construction processing on multiple single-frame multi-source activity environment data along the time axis to form continuous spatiotemporal sequence samples. Continuous sequence construction processing refers to selecting multiple single-frame multi-source activity environment data consecutively according to time unit order and organizing them into a spatiotemporal sequence that the model can process. Continuous spatiotemporal sequence samples are the data samples used as input to the ConvLSTM-PINN prediction model, capable of simultaneously expressing historical changes in fishing vessel activity intensity, supplementary information on covert operations, and changes in the marine environment. The server sets the historical observation window length and the prediction window length, using single-frame multi-source activity environment data corresponding to multiple consecutive time units as historical sequences, and using the target fishing vessel activity intensity field following the historical sequences as prediction labels. The expression for constructing continuous spatiotemporal sequence samples is:

[0138]

[0139]

[0140] in, Indicates the first Historical sequences in a continuous spatiotemporal sequence sample constructed starting from a time unit; Indicates the first Single-frame multi-source activity environment data corresponding to each time unit; Indicates the first Single-frame multi-source activity environment data corresponding to each time unit; This represents the single-frame multi-source activity environment data corresponding to the last time unit in the historical sequence; This indicates the length of the historical observation window, which is set by the system based on the cycle of fishing vessel activity and model input requirements. Indicates the first Predictive labels are constructed starting from each time unit; This represents the target fishing vessel activity intensity field corresponding to the first time unit after the historical sequence. This indicates the prediction window length, which is set by the system according to the prediction task requirements. The above expression constructs historical sequences and prediction labels by sliding along the time axis, enabling continuous spatiotemporal sequence samples to be used to train or drive the ConvLSTM-PINN prediction model, thereby generating subsequent prediction results of fishing vessel activity intensity based on historical multi-source activity environment changes.

[0141] S160. Input the continuous spatiotemporal sequence samples into the ConvLSTM-PINN prediction model to generate the prediction results of fishing vessel activity intensity based on the physical constraint mechanism constructed by ocean current constraint information, spatial continuity constraint information, wind field safety constraint information and biological coupling constraint information.

[0142] Specifically, the server first acquires the target fishing vessel activity intensity field, wind field status, ocean current status, and chlorophyll concentration status from continuous spatiotemporal sequence samples, and constructs a historical activity environment sequence based on the historical observation window length. The target fishing vessel activity intensity field is spatial distribution data formed by fusing AIS fishing activity intensity and covert operation activity intensity, used to characterize the strength of fishing vessel activities in the target sea area; the wind field status is used to characterize the impact of wind speed and direction on the safety of fishing vessel operations in the target sea area; the ocean current status is used to characterize the impact of seawater flow on nutrients and fish aggregation in the target sea area; and the chlorophyll concentration status is used to characterize the impact of primary productivity distribution on fish resource enrichment in the target sea area. The server selects multi-source activity environment data corresponding to multiple consecutive time units from the continuous spatiotemporal sequence samples according to the chronological order of time units, and organizes them into a historical activity environment sequence, enabling the ConvLSTM-PINN prediction model to simultaneously read historical changes in fishing vessel activities and historical changes in the marine environment. The expression for the historical activity environment sequence is:

[0143]

[0144] in, Indicates the first A sequence of historical activity environments constructed starting from each time unit; Indicates the first The single-frame multi-source activity environment data corresponding to each time unit is obtained by combining the target fishing vessel activity intensity field, wind field status, ocean current status, and chlorophyll concentration status channels. Indicates the first Single-frame multi-source activity environment data corresponding to each time unit; This represents the single-frame multi-source activity environment data corresponding to the last time unit in the historical activity environment sequence; This indicates the length of the historical observation window, which is set by the system based on the cycle of fishing vessel activity, the cycle of marine environmental change, and the length of the model input. This indicates the starting time unit number of the historical activity environment sequence. This expression, through the splicing of continuous time units, enables the model to obtain a continuous process of the change in the target fishing vessel activity intensity field and the marine environmental state over time, rather than just obtaining a static distribution at a single moment.

[0145] The server inputs the historical activity environment sequence into the spatiotemporal feature extraction structure of the ConvLSTM-PINN prediction model. The ConvLSTM-PINN prediction model is a prediction model formed by combining a convolutional long short-term memory network (LSTM) and a physical information neural network (PIN) constraint mechanism. The LSTM network is used to extract spatial distribution features and temporal evolution features, while the PSN constraint mechanism incorporates ocean physical laws into the prediction process. The spatiotemporal feature extraction structure reads the historical activity environment sequence frame by frame in chronological order, and within each time unit, extracts the spatial neighborhood relationships between the target fishing vessel's activity intensity field, wind field state, ocean current state, and chlorophyll concentration state through convolutional operations. Simultaneously, it preserves the evolutionary relationships between adjacent time units through gated memory updates. The state update expression of the spatiotemporal feature extraction structure can be represented as:

[0146]

[0147] in, Indicates the first Spatiotemporal implicit features output by each time unit; Indicates the first The server reads the multi-source activity environment data of a single frame corresponding to each time unit from the historical activity environment sequence; Indicates the first The historical implicit characteristics passed from one time unit to the current time unit; This represents the convolutional long short-term memory processing function, used to extract temporal dependencies while preserving spatial structure; This represents the time unit number in the historical activity environment sequence. This expression combines the data of the current time unit with the implicit features of the previous time unit, so that the spatiotemporal implicit features simultaneously include the current spatial distribution of the sea area and historical temporal evolution information.

[0148] The server then performs spatial location enhancement and channel contribution adjustment on the spatiotemporal latent features using an attention enhancement structure. Spatial location enhancement identifies spatial regions in the spatiotemporal latent features that are more relevant to fishing vessel activity hotspots and increases the expression intensity of these regions in subsequent predictions. Channel contribution adjustment identifies feature channels that contribute significantly to the prediction results from the target fishing vessel activity intensity field, wind field state, ocean current state, and chlorophyll concentration state, and increases the contribution of effective channels while reducing the contribution of noisy channels. The server can first perform channel compression and spatial compression on the spatiotemporal latent features to obtain channel attention weights and spatial attention weights, and then use these weights to weight the spatiotemporal latent features, forming enhanced spatiotemporal latent features. The expression for enhanced spatiotemporal latent features is:

[0149]

[0150] in, This indicates enhanced spatiotemporal implicit features; This represents the spatiotemporal latent features output by the spatiotemporal feature extraction structure; The channel attention weight is calculated by the server based on the global response intensity of each channel in the spatiotemporal latent features, and is used to characterize the contribution of different feature channels to the prediction of fishing vessel activities. The spatial attention weight is calculated by the server based on the local response intensity of each spatial location in the spatiotemporal implicit features, and is used to characterize the degree of contribution of different spatial locations to the prediction of fishing vessel activities. This indicates element-wise multiplication. The expression first uses channel attention weights to adjust the contributions of different feature channels, and then uses spatial attention weights to highlight potential fishing vessel activity areas, thus reducing the model's dependence on irrelevant background sea areas and weakly correlated environmental channels.

[0151] The server performs activity intensity decoding on the enhanced spatiotemporal latent features using an activity intensity decoding structure. This structure transforms the high-dimensional latent features into a network structure that converts them into a spatial distribution of fishing vessel activity intensity. It is used to restore the enhanced spatiotemporal latent features to spatial prediction results corresponding to future time units. The server performs convolutional mapping on the enhanced spatiotemporal latent features, converting them from multi-channel feature maps to single-channel activity intensity prediction maps, and applies non-negativity constraints to the decoding results. These constraints ensure that the predicted fishing vessel activity intensity does not have negative values, as activity intensity represents the number, density, or contribution of activities and cannot be negative. The initial expression for the fishing vessel activity intensity prediction field is:

[0152]

[0153] in, This represents the initial predicted field of fishing vessel activity intensity; This indicates enhanced spatiotemporal implicit features; This represents a convolutional decoding function used to map enhanced spatiotemporal latent features into activity intensity prediction maps corresponding to future time units; This represents a non-negative activation function used to convert the convolution decoding result into a non-negative value. This expression obtains the future spatial distribution of fishing vessel activity through convolution decoding, and uses the non-negative activation function to ensure that the activity intensity prediction result conforms to the physical meaning of activity intensity.

[0154] The expression for the nonnegative activation function is:

[0155] in, This represents the original predicted value corresponding to any spatial unit output by the convolutional decoding function; It represents exponential operations with the natural constant as the base; This represents the predicted value after non-negative activation. This function... It outputs a positive value for any real number and has continuous differentiability compared to directly truncating negative values, which is beneficial for gradient updates during model training.

[0156] The server constructs ocean current constraint information based on the initial predicted field of fishing vessel activity intensity and ocean current conditions. This constraint information is used to limit the model's output of high-intensity fishing vessel activity predictions in ocean current divergence areas. Ocean current conditions typically include east-west and north-south current components. The server determines ocean current divergence by calculating the degree of change in ocean current conditions in spatial directions. When ocean current divergence is positive, it indicates a divergence trend near the corresponding spatial unit, making it difficult for nutrients and plankton to accumulate, and generally preventing the formation of stable, high-intensity fishing hotspots. When ocean current divergence is negative, it indicates a convergence trend near the corresponding spatial unit, which is more consistent with the conditions for fish resource enrichment. The expression for ocean current constraint information is:

[0157]

[0158] in, Indicates ocean current constraint information; This represents the total number of spatial units in a unified spatial grid, obtained by dividing the target sea area. Indicates the spatial unit number; Indicating the initial prediction field of fishing vessel activity intensity, the [missing information] is... Predicted values ​​for each spatial unit; Indicates the first The ocean current state corresponding to each spatial unit is obtained from the ocean current data of future time units through spatial alignment and temporal alignment processing; Indicates the first The current divergence corresponding to each spatial unit is calculated by the difference of the current state in the spatial direction; This represents a linear rectification function that retains the input value when it is positive and outputs zero when the input value is less than or equal to zero. This expression penalizes high activity intensity predictions occurring in current divergence areas by multiplying the initial predicted fishing vessel activity intensity by the current divergence degree, thus making the prediction results more consistent with the relationship between current convergence and resource enrichment.

[0159] The server constructs spatial continuity constraint information based on the initial predicted field of fishing vessel activity intensity. This spatial continuity constraint information is used to ensure the prediction results maintain reasonable spatial continuity, avoiding isolated high-value hotspots due to sample sparsity, false detections at night, or local noise. The server calculates the variation amplitude of the initial predicted field of fishing vessel activity intensity between adjacent spatial cells and sums the variation amplitudes in the horizontal and vertical directions to form the spatial continuity constraint information. The expression for the spatial continuity constraint information is:

[0160]

[0161] in, Represents spatial continuity constraint information; This represents the total number of spatial units in a unified spatial grid; Indicates the spatial unit number; Indicating the initial prediction field of fishing vessel activity intensity, the [missing information] is... Predicted values ​​for each spatial unit; Indicates the first The difference in predicted values ​​between a spatial cell and its neighboring spatial cells along the longitude direction is obtained by the difference in predicted values ​​between the neighboring spatial cells. Indicates the first The difference in predicted values ​​between a spatial cell and its neighboring spatial cells in the latitudinal direction is obtained by the difference in predicted values ​​between the neighboring spatial cells. This expression represents absolute value operations. By limiting abrupt changes in predicted values ​​between adjacent spatial cells, it makes the spatial distribution of fishing vessel activity intensity more continuous and reduces the impact of isolated noise hotspots on the prediction results.

[0162] The server constructs wind field safety constraint information based on the initial predicted fishing vessel activity intensity field and wind field conditions. This constraint information is used to limit the model's output of fishing vessel activity predictions in high-wind-speed areas unsuitable for operations. Wind field conditions typically include east-west and north-south wind field components. The server calculates the wind speed modulus based on these two components and compares it to a preset safe wind speed threshold. When the wind speed modulus corresponding to a spatial cell exceeds the preset safe wind speed threshold, it indicates that the sea state in that area is unsuitable for fishing vessel operations. If the initial predicted fishing vessel activity intensity field still outputs a high prediction value for that spatial cell, then constraints should be imposed. The expression for the wind field safety constraint information is:

[0163]

[0164] in, This indicates wind farm safety constraint information; This represents the total number of spatial units in a unified spatial grid; Indicates the spatial unit number; Indicating the initial prediction field of fishing vessel activity intensity, the [missing information] is... Predicted values ​​for each spatial unit; Indicates the first The wind field state corresponding to each spatial unit is obtained from the wind field data of the future time unit through spatial alignment and temporal alignment processing; Indicates the first The wind speed modulus corresponding to each spatial unit is calculated from the east-west wind field component and the north-south wind field component. This indicates a preset safe wind speed threshold, set by fishing vessel safety operation rules or regulatory requirements. This indicates an indicator function, which takes the value 1 when the condition within the parentheses is true, and 0 otherwise. This expression extracts the predicted activity intensity only from spatial cells where the wind speed modulus reaches or exceeds a preset safe wind speed threshold, and imposes constraints on these predicted values, enabling the model to reduce unreasonable operation predictions under severe wind conditions.

[0165] The expression for the wind speed modulus is:

[0166] in, Indicates the first Wind speed modulus corresponding to each spatial unit; Indicates the first The server reads the east-west wind field components corresponding to each spatial unit from the wind field status. Indicates the first The server reads the north-south wind field components corresponding to each spatial unit from the wind field status. This expression synthesizes the wind speed intensity by combining the wind field components in two orthogonal directions, and is used to determine whether the spatial unit exceeds a preset safe wind speed threshold.

[0167] The server constructs biocoupled constraint information based on the initial predicted field of fishing vessel activity intensity and chlorophyll concentration status. This biocoupled constraint information is used to maintain a reasonable spatial correspondence between high-value areas of fishing vessel activity and areas with high chlorophyll concentration. Chlorophyll concentration status is typically used to characterize the level of marine primary productivity; higher chlorophyll concentration often indicates abundant phytoplankton, which may further attract fish populations. Therefore, fishing vessel activity hotspots should statistically maintain a certain coupling with areas of higher chlorophyll concentration. The server normalizes both the initial predicted field of fishing vessel activity intensity and the chlorophyll concentration status, and calculates their overlap within the same spatial unit. Since the model optimization objective is usually to reduce constraint values, a negative overlap term guides the model to enhance the consistency between activity intensity prediction and chlorophyll concentration status. The expression for the biocoupled constraint information is:

[0168]

[0169] in, Represents biological coupling constraint information; This represents the total number of spatial units in a unified spatial grid; Indicates the spatial unit number; This represents the normalized initial predicted value of fishing vessel activity intensity, which is obtained by normalizing the predicted values ​​in the initial fishing vessel activity intensity prediction field. This represents the normalized chlorophyll concentration state, obtained from chlorophyll concentration data of future time units through spatial alignment, temporal alignment, and normalization. This expression, by calculating the spatial overlap between normalized activity intensity and normalized chlorophyll concentration, guides the prediction of fishing vessel activity intensity to be more inclined towards areas with higher primary productivity.

[0170] The server constructs a physical constraint mechanism based on ocean current constraints, spatial continuity constraints, wind field safety constraints, and biological coupling constraints. This physical constraint mechanism is then used to perform constraint optimization on the initial predicted fishing vessel activity intensity field. The physical constraint mechanism refers to a constraint system that integrates ocean current convergence laws, spatial continuity laws, wind field safety laws, and biological coupling laws into the model prediction process. It can influence model parameters inversely through the loss function during model training, and it can also perform post-processing corrections on the initial predicted fishing vessel activity intensity field during model inference. The expression for the physical constraint mechanism is:

[0171]

[0172] in, This represents the comprehensive physical constraint value corresponding to the physical constraint mechanism; Indicates ocean current constraint information; Represents spatial continuity constraint information; This indicates wind farm safety constraint information; Represents biological coupling constraint information; This represents the current constraint adjustment coefficient, used to control the contribution of current constraint information to the overall physical constraint value; This represents the spatial continuity constraint adjustment coefficient, used to control the contribution of spatial continuity constraint information to the overall physical constraint value; This represents the wind field safety constraint adjustment coefficient, used to control the contribution of wind field safety constraint information to the overall physical constraint value; This represents the biological coupling constraint adjustment coefficient, used to control the contribution of biological coupling constraint information to the overall physical constraint value. This adjustment coefficient can be set by the server based on prediction errors on the training and validation sets, the degree of physical violation, and regulatory application requirements. This expression transforms different physical laws into constraint values ​​that can be jointly optimized, ensuring that the model output not only conforms to historical data patterns but is also constrained by objective marine environmental conditions.

[0173] When training the ConvLSTM-PINN prediction model, the server can combine the data fitting error and the comprehensive physical constraint value to form the overall optimization objective. The data fitting error measures the difference between the initial predicted fishing vessel activity intensity field and the actual target fishing vessel activity intensity field, while the comprehensive physical constraint value measures the degree to which the initial predicted fishing vessel activity intensity field violates oceanographic laws. The expression for the overall optimization objective is:

[0174]

[0175] in, This represents the overall optimization objective of the ConvLSTM-PINN prediction model; This represents the data fitting error, calculated from the difference between the initial predicted field of fishing vessel activity intensity and the actual target fishing vessel activity intensity field in the training samples; This represents the physical constraint strength coefficient, which is used to control the degree of influence of the comprehensive physical constraint value on model training. This represents the comprehensive physical constraint value. This expression reduces both prediction error and physical violation, enabling the model to learn historical fishing vessel activity patterns while avoiding predictions that significantly conflict with ocean currents, spatial continuity, wind field safety, and chlorophyll distribution.

[0176] The ConvLSTM-PINN prediction model first uses historical activity environmental sequences as training samples, organizing the target fishing vessel activity intensity field, wind field state, ocean current state, and chlorophyll concentration state into continuous spatiotemporal input data in chronological order. The ConvLSTM network extracts the spatiotemporal correlation features between changes in fishing vessel activity and changes in the marine environment, and an attention enhancement mechanism is used to highlight the contributions of key activity areas and key environmental factors. Subsequently, an activity intensity decoding structure generates the initial fishing vessel activity intensity prediction field for future times. During model training, a PINN physical constraint mechanism is introduced. Ocean current constraint information is constructed based on the initial fishing vessel activity intensity prediction field and ocean current state; spatial continuity constraint information is constructed based on the initial fishing vessel activity intensity prediction field; wind field safety constraint information is constructed based on the initial fishing vessel activity intensity prediction field and wind field state; and biological coupling constraint information is constructed based on the initial fishing vessel activity intensity prediction field and chlorophyll concentration state. These constraints collectively form the physical constraint mechanism. During training, both prediction error and physical constraint error are optimized, enabling the model to learn historical fishing vessel activity patterns while also satisfying ocean current convergence patterns, spatial continuity patterns, operational safety patterns, and marine ecological patterns. After training, the constructed ConvLSTM-PINN prediction model is used to predict historical activity environmental sequences, outputting a target fishing vessel activity intensity prediction field that conforms to the laws of ocean physics, and generating prediction results such as the spatial distribution of fishing vessel activity intensity, activity hotspot areas, and low activity intensity areas.

[0177] The server uses a physical constraint mechanism to perform constraint optimization on the initial fishing vessel activity intensity prediction field, forming the target fishing vessel activity intensity prediction field. The target fishing vessel activity intensity prediction field is the prediction result after the initial fishing vessel activity intensity prediction field has been adjusted by the physical constraint mechanism. Compared to the initial prediction field, its abnormally high values ​​in the current divergence area are weakened, unreasonable activity predictions in high wind speed areas are suppressed, spatially isolated abrupt prediction areas are smoothed, and a more reasonable activity response is maintained in chlorophyll-rich areas suitable for biological resource enrichment. The server generates fishing vessel activity intensity prediction results based on the target fishing vessel activity intensity prediction field. These results include the spatial distribution of fishing vessel activity intensity, fishing vessel activity hotspots, low activity intensity areas, and physical constraint adjustment records. The spatial distribution of fishing vessel activity intensity is used to represent the intensity of fishing vessel activity in each spatial unit of the future target sea area. Fishing vessel activity hotspots represent spatial areas where the predicted value is higher than the hotspot determination threshold, and low activity intensity areas represent spatial areas where the predicted value is lower than the low activity determination threshold. Physical constraint adjustment records record the spatial locations and reasons for adjustments made to the prediction results by ocean current constraints, spatial continuity constraints, wind field safety constraints, and biological coupling constraints. The determination expressions for hotspots and low activity areas are:

[0178]

[0179] in, Indicates the first The hotspot region marker corresponding to each spatial unit, when At that time, the server will... Each spatial unit was identified as a hotspot for fishing vessel activity. Indicates the first The low-activity region marker corresponding to each spatial unit, when At that time, the server will... Each spatial unit was identified as a low-activity area; Indicating the target fishing vessel activity intensity prediction field, the first... Predicted values ​​for each spatial unit; The threshold for identifying hotspots is determined by the activity intensity quantile of historically high-intensity work areas or by regulatory requirements. The threshold for determining low activity is determined by the activity intensity quantile of historically low-intensity areas or by regulatory requirements. This indicates the indicator function. The expression uses a threshold determination to transform the continuous prediction field into spatial area results that can be used for regulatory visualization and risk analysis, thereby facilitating subsequent zoning management of potentially high-activity and low-activity areas within the target sea area.

[0180] This application also provides a device for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints, referring to... Figure 2 , Figure 2This is a flowchart illustrating the fishing vessel activity identification and prediction method based on multi-source data fusion and physical constraints provided in this application embodiment. The device is a server, which includes an acquisition module 21 and a processing module 22. The acquisition module 21 is used to acquire AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data in the target sea area. The processing module 22 is used to extract ground speed features and heading change features from the AIS trajectory data, and perform time-series delay embedding processing based on the ground speed features and heading change features to construct a time-series state vector. The processing module 22 is also used to perform cluster analysis on the time-series state vector to form fishing state categories and navigation state categories, and perform state labeling on the AIS trajectory data based on the fishing state categories and navigation state categories to form fishing vessel activity trajectories. The processing module 22 is also used to perform spatiotemporal cross-matching processing on VIIRS / DNB nighttime light remote sensing data and fishing vessel activity trajectory data to identify unmatched nighttime light targets and identify concealed nighttime operation targets based on the radiance distribution of unmatched nighttime light targets, so as to form abnormal fishing vessel activity identification results; the processing module 22 is also used to construct a fishing vessel activity intensity field based on the abnormal fishing vessel activity identification results and fishing vessel activity trajectory data corresponding to the fishing status category, and combine it with marine environmental data to form a continuous spatiotemporal sequence sample; the processing module 22 is also used to input the continuous spatiotemporal sequence sample into the ConvLSTM-PINN prediction model to generate fishing vessel activity intensity prediction results based on the physical constraint mechanism constructed by ocean current constraint information, spatial continuity constraint information, wind field safety constraint information and biological coupling constraint information.

[0181] This application also provides an electronic device, with reference to... Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 31, at least one network interface 34, a user interface 33, a memory 35, and at least one communication bus 32.

[0182] The communication bus 32 is used to enable communication between these components.

[0183] The user interface 33 may include a display screen and a camera, and may also include standard wired and wireless interfaces.

[0184] The network interface 34 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0185] The processor 31 may include one or more processing cores. The processor 31 connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in the memory 35, and calling data stored in the memory 35 to perform various server functions and process data. Optionally, the processor 31 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 31 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 31 and may be implemented as a separate chip.

[0186] The memory 35 may include random access memory (RAM) or read-only memory. Optionally, the memory 35 may include a non-transitory computer-readable storage medium. The memory 35 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 35 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 35 may also be at least one storage device located remotely from the aforementioned processor 31. Figure 3 As shown, the memory 35, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints.

[0187] exist Figure 3In the electronic device shown, the user interface 33 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 31 can be used to call the application program stored in the memory 35 that is based on the fishing vessel activity identification and prediction method based on multi-source data fusion and physical constraints. When executed by one or more processors, the electronic device executes one or more methods as described in the above embodiments.

[0188] This application also provides a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0189] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A fishing boat activity recognition and prediction method based on multi-source data fusion and physical constraints, characterized in that, The method includes: Acquire AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data in the target sea area; Based on the AIS trajectory data, ground speed features and heading change features are extracted, and time-series delay embedding processing is performed based on the ground speed features and heading change features to construct a time-series state vector. Cluster analysis is performed on the time-series state vector to form fishing state categories and navigation state categories, and state annotation is performed on the AIS trajectory data based on the fishing state categories and the navigation state categories to form fishing vessel activity trajectory data; Spatiotemporal cross-matching processing is performed on the VIIRS / DNB nighttime light remote sensing data and the fishing vessel activity trajectory data to identify unmatched nighttime light targets, and the nighttime concealed operation targets are identified based on the radiance distribution of the unmatched nighttime light targets to form abnormal fishing vessel activity identification results. Based on the abnormal fishing vessel activity identification results and the fishing vessel activity trajectory data corresponding to the fishing status category, a fishing vessel activity intensity field is constructed, and combined with the marine environmental data to form a continuous spatiotemporal sequence sample. The continuous spatiotemporal sequence samples are input into the ConvLSTM-PINN prediction model to generate prediction results of fishing vessel activity intensity based on the physical constraint mechanism constructed from ocean current constraint information, spatial continuity constraint information, wind field safety constraint information, and biological coupling constraint information.

2. The fishing boat activity recognition and prediction method based on multi-source data fusion and physical constraints according to claim 1, characterized in that, The acquisition of AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data in the target sea area specifically includes: The original AIS trajectory data in the target sea area is obtained, and fishing vessel trajectory points are filtered according to vessel type information. At the same time, the fishing vessel trajectory points are assigned according to vessel identification, and a continuous AIS trajectory sequence is formed according to trajectory time. The continuous AIS trajectory sequence is validated to remove abnormal trajectory points, and missing trajectory points are filled in according to the time interval between adjacent trajectory points to form target AIS trajectory data. The original VIIRS / DNB nighttime light remote sensing data in the target sea area are acquired, and invalid nighttime light detection points are removed according to spatial range, target time period and quality label to form a target nighttime light detection point set; The raw marine environmental data in the target sea area is acquired. The raw marine environmental data includes wind field data, ocean current data, and chlorophyll concentration data. Based on the spatial range and the target time period, the wind field data, ocean current data, and chlorophyll concentration data are cropped and missing data are filled in to form an environmental data set. Based on a unified spatiotemporal reference, spatiotemporal alignment processing is performed on the target AIS trajectory data, the target night light detection point set, and the environmental data set, and a data association index is established to form AIS trajectory data, VIIRS / DNB night light remote sensing data, and marine environmental data. 3.The fishing boat activity recognition and prediction method based on multi-source data fusion and physical constraints according to claim 2, characterized in that, The step of extracting ground speed features and heading change features from the AIS trajectory data, and performing time-delay embedding processing based on the ground speed features and heading change features to construct a time-series state vector, specifically includes: Ground speed features are extracted from the ground speed in the continuous AIS trajectory sequence, and abnormal ground speeds are corrected to form target ground speed features. The degree of heading change between adjacent trajectory points is extracted based on the heading angle in the continuous AIS trajectory sequence, and the heading change feature is formed by combining the time interval between the corresponding trajectory times. The target's ground speed characteristics and heading change characteristics are correlated and combined to form a sequence of trajectory point motion characteristics; The motion feature sequence of the trajectory points is processed by a sliding window according to the preset embedding window length, and the motion features of the trajectory points corresponding to multiple adjacent trajectory points are extracted continuously and then spliced ​​according to the trajectory time order to form the corresponding temporal state vector.

4. The method of claim 1, wherein, The process of performing cluster analysis on the time-series state vector to form fishing state categories and navigation state categories, and then performing state annotation on the AIS trajectory data based on the fishing state categories and navigation state categories to form fishing vessel activity trajectory data, specifically includes: Obtain the ship identifier, start trajectory time, end trajectory time, and associated trajectory point position corresponding to the time-series state vector, and perform feature scale unification processing on the time-series state vector; Cluster analysis is performed on the time-series state vector that has completed feature scale unification processing based on the preset number of cluster categories. Multiple stable categories are formed by iteratively performing category assignment processing and category center update processing. The fishing status category and navigation status category are determined based on the ground speed characteristic distribution and heading change characteristic distribution corresponding to each stability category; A status label is generated based on the fishing status category and the navigation status category, and the status label is written back to the trajectory point corresponding to the AIS trajectory data to form an initial status labeling result; The initial state labeling results are subjected to state consistency verification, and isolated state segments are corrected according to the state labels corresponding to adjacent trajectory points to form the final state labeling results. Fishing vessel activity trajectory data is generated based on the final status annotation results. The fishing vessel activity trajectory data includes vessel identification, trajectory time, latitude and longitude position, ground speed characteristics, heading change characteristics, status label, and trajectory interruption mark.

5. The method of claim 1, wherein, The process of performing spatiotemporal cross-matching processing on the VIIRS / DNB nighttime light remote sensing data and the fishing vessel activity trajectory data to identify unmatched nighttime light targets, and identifying concealed nighttime operation targets based on the radiance distribution of the unmatched nighttime light targets, in order to form an abnormal fishing vessel activity identification result, specifically includes: The VIIRS / DNB night light remote sensing data is subjected to quality screening processing to remove invalid night light detection points and retain night light detection points that meet preset conditions to form candidate night light targets; A trajectory matching index is constructed based on the fishing vessel activity trajectory data, and the trajectory silence interval is determined based on the trajectory interruption marker; Spatiotemporal cross-matching is performed on the candidate night-light targets and the trajectory matching index to form matched night-light targets and unmatched night-light targets; Based on the spatiotemporal relationship between the unmatched night-light targets and the trajectory silence interval, a silent association judgment is performed to form silent associated night-light targets and independent unmatched night-light targets; A radiance distribution is constructed based on the radiance values ​​of the unmatched night-light targets, and a radiance segmentation threshold is determined based on the radiance distribution. The unmatched night-light targets are then divided into low-brightness unmatched targets and high-brightness unmatched targets based on the radiance segmentation threshold. Identify concealed nighttime operation targets based on the high-brightness unmatched targets and the silent associated nighttime luminous targets, and generate abnormal fishing vessel activity identification results based on the concealed nighttime operation targets. The abnormal fishing vessel activity identification results include nighttime luminous detection time, nighttime luminous detection location, radiance value, silent association status, and target type.

6. The method of claim 1, wherein, The process of constructing a fishing vessel activity intensity field based on the abnormal fishing vessel activity identification results and the fishing vessel activity trajectory data corresponding to the fishing status category, and combining it with the marine environmental data to form a continuous spatiotemporal sequence sample, specifically includes: Establish a unified spatial grid and a unified time step covering the target sea area, and perform spatial mapping and temporal mapping processing on the fishing vessel activity trajectory data corresponding to the fishing status category based on the unified spatial grid and the unified time step to form the AIS fishing activity intensity. Based on the unified spatial grid and the unified time step, spatial mapping and temporal mapping processing are performed on the identification results of the abnormal fishing vessel activities, and the intensity of the corresponding covert operation activities is determined according to the radiance value, silent association state and target type. The intensity of AIS fishing activities and the intensity of covert operations are fused together, and the correlation between the intensity of covert operations and the intensity of AIS fishing activities is determined based on the trajectory interruption marker and the silent association state, so as to form a fishing vessel activity intensity field. The activity intensity field of the fishing vessels is subjected to continuous temporal and spatial verification, and the abnormal activity intensity is corrected according to the activity intensity distribution of fishing vessels corresponding to adjacent time units and adjacent spatial units to form the target fishing vessel activity intensity field. Based on the unified spatial grid and the unified time step, the marine environmental data is subjected to spatial alignment and temporal alignment processing to form wind field state, ocean current state and chlorophyll concentration state corresponding to the target fishing vessel activity intensity field; The target fishing vessel activity intensity field, wind field state, ocean current state, and chlorophyll concentration state are combined into channels to form single-frame multi-source activity environment data. Continuous sequence construction processing is then performed on multiple single-frame multi-source activity environment data along the time axis to form continuous spatiotemporal sequence samples.

7. The method for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints according to claim 1, characterized in that, The step of inputting the continuous spatiotemporal sequence samples into the ConvLSTM-PINN prediction model to generate fishing vessel activity intensity prediction results based on a physical constraint mechanism constructed from ocean current constraint information, spatial continuity constraint information, wind field safety constraint information, and biological coupling constraint information specifically includes: The activity intensity field, wind field state, ocean current state, and chlorophyll concentration state of the target fishing vessel in the continuous spatiotemporal sequence sample are obtained, and a historical activity environment sequence is constructed based on the length of the historical observation window. The spatiotemporal feature extraction structure in the ConvLSTM-PINN prediction model is used to perform spatiotemporal feature extraction processing on the historical activity environment sequence to form spatiotemporal implicit features characterizing the evolution process of fishing vessel activities and the process of marine environmental change. The attention enhancement structure in the ConvLSTM-PINN prediction model is used to perform spatial location enhancement and channel contribution adjustment on the spatiotemporal latent features to form enhanced spatiotemporal latent features. The activity intensity decoding structure in the ConvLSTM-PINN prediction model is used to perform activity intensity decoding processing on the enhanced spatiotemporal latent features, and non-negative constraint processing is performed on the decoding results to form an initial fishing vessel activity intensity prediction field. Ocean current constraint information is constructed based on the initial predicted field of fishing vessel activity intensity and the ocean current state; spatial continuity constraint information is constructed based on the initial predicted field of fishing vessel activity intensity; wind field safety constraint information is constructed based on the initial predicted field of fishing vessel activity intensity and the wind field state; and biological coupling constraint information is constructed based on the initial predicted field of fishing vessel activity intensity and the chlorophyll concentration state. A physical constraint mechanism is constructed based on the ocean current constraint information, the spatial continuity constraint information, the wind field safety constraint information, and the biological coupling constraint information. The physical constraint mechanism is then used to perform constraint optimization processing on the initial fishing vessel activity intensity prediction field to form the target fishing vessel activity intensity prediction field. Based on the target fishing vessel activity intensity prediction field, a fishing vessel activity intensity prediction result is generated. The fishing vessel activity intensity prediction result includes the spatial distribution of fishing vessel activity intensity, fishing vessel activity hotspot areas, low activity intensity areas, and physical constraint adjustment records.

8. A device for identifying and predicting fishing vessel activities based on multi-source data fusion and physical constraints, characterized in that, The apparatus is used to execute the fishing vessel activity identification and prediction method based on multi-source data fusion and physical constraints as described in any one of claims 1 to 7. The apparatus includes an acquisition module and a processing module, wherein... The acquisition module is used to acquire AIS trajectory data, VIIRS / DNB nighttime light remote sensing data, and marine environmental data in the target sea area; The processing module is used to extract ground speed features and heading change features from the AIS trajectory data, and to perform time delay embedding processing based on the ground speed features and heading change features to construct a time-series state vector. The processing module is further configured to perform cluster analysis on the time-series state vector to form fishing state categories and navigation state categories, and to perform state annotation on the AIS trajectory data based on the fishing state categories and the navigation state categories to form fishing vessel activity trajectory data. The processing module is also used to perform spatiotemporal cross-matching processing on the VIIRS / DNB night light remote sensing data and the fishing vessel activity trajectory data to determine unmatched night light targets, and to identify concealed nighttime operation targets based on the radiance distribution of the unmatched night light targets, so as to form abnormal fishing vessel activity identification results. The processing module is also used to construct a fishing vessel activity intensity field based on the abnormal fishing vessel activity identification results and the fishing vessel activity trajectory data corresponding to the fishing status category, and to form a continuous spatiotemporal sequence sample by combining the marine environmental data. The processing module is also used to input the continuous spatiotemporal sequence samples into the ConvLSTM-PINN prediction model to generate prediction results of fishing vessel activity intensity based on the physical constraint mechanism constructed by ocean current constraint information, spatial continuity constraint information, wind field safety constraint information and biological coupling constraint information.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the fishing vessel activity identification and prediction method based on multi-source data fusion and physical constraints as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores instructions that, when executed, perform the fishing vessel activity identification and prediction method based on multi-source data fusion and physical constraints as described in any one of claims 1 to 7.

Images